frcnn_train_vgg.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "frcnn_train_vgg.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Vram97/Car-and-Truck-classification/blob/main/frcnn_train_vgg.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eV4T9G4LCH8t"
      },
      "source": [
        "from google.colab import drive"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_fzxugFjqc2T",
        "outputId": "4e5da341-f839-4bde-c88b-935fad159871"
      },
      "source": [
        "drive.mount('/gdrive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /gdrive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "J2EBHYjgRs1G",
        "outputId": "dc32bb5c-98ec-4d74-9fbd-349377ef21f1"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/drive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "D7zwItc4cYHm",
        "outputId": "d03ef8cc-b78c-4b77-8def-fc2e0e1cff43"
      },
      "source": [
        "! unzip drive/My\\Drive/CNNDataset.zip"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Archive:  drive/MyDrive/CNNDataset.zip\n",
            "   creating: CNNDataset/\n",
            "  inflating: CNNDataset/train-annotations-bbox.csv  \n",
            "  inflating: CNNDataset/train-images-boxable-with-rotation.csv  \n",
            "  inflating: CNNDataset/class-descriptions-boxable.csv  \n",
            "  inflating: CNNDataset/train.csv    \n",
            "  inflating: CNNDataset/test.csv     \n",
            "  inflating: CNNDataset/test_annotation.txt  \n",
            "  inflating: CNNDataset/annotation.txt  \n",
            "   creating: CNNDataset/Truck/\n",
            "  inflating: CNNDataset/Truck/9001036c66a19a83.jpg  \n",
            "  inflating: CNNDataset/Truck/7cc6e09f8f1d2698.jpg  \n",
            "  inflating: CNNDataset/Truck/85151615a3dd7394.jpg  \n",
            "  inflating: CNNDataset/Truck/e1dc0a2765038997.jpg  \n",
            "  inflating: CNNDataset/Truck/faf8854c355a752a.jpg  \n",
            "  inflating: CNNDataset/Truck/0786212bde3d2c9c.jpg  \n",
            "  inflating: CNNDataset/Truck/00e577cb0dd95bf4.jpg  \n",
            "  inflating: CNNDataset/Truck/df6cfb5e5f9f967d.jpg  \n",
            "  inflating: CNNDataset/Truck/21f283bece63c7c9.jpg  \n",
            "  inflating: CNNDataset/Truck/bff97f2b287f543f.jpg  \n",
            "  inflating: CNNDataset/Truck/ea87ffff22c14e7a.jpg  \n",
            "  inflating: CNNDataset/Truck/e8a2a51e75297966.jpg  \n",
            "  inflating: CNNDataset/Truck/06ab38835a7a8a58.jpg  \n",
            "  inflating: CNNDataset/Truck/606e8d152ae2c704.jpg  \n",
            "  inflating: CNNDataset/Truck/31c0bf8c89fca5e5.jpg  \n",
            "  inflating: CNNDataset/Truck/582df91290740cfa.jpg  \n",
            "  inflating: CNNDataset/Truck/0a5b2310fe6e429f.jpg  \n",
            "  inflating: CNNDataset/Truck/71cc934013f02bfe.jpg  \n",
            "  inflating: CNNDataset/Truck/89eb16acf43df012.jpg  \n",
            "  inflating: CNNDataset/Truck/55d9de60ad6279c5.jpg  \n",
            "  inflating: CNNDataset/Truck/dac8ac5ad2bca314.jpg  \n",
            "  inflating: CNNDataset/Truck/b82b8ddb96dd6f3a.jpg  \n",
            "  inflating: CNNDataset/Truck/b1a122749844023d.jpg  \n",
            "  inflating: CNNDataset/Truck/f5ff3013e4f0b1d1.jpg  \n",
            "  inflating: CNNDataset/Truck/a709f87623a9e4b9.jpg  \n",
            "  inflating: CNNDataset/Truck/0b8f012362e805c9.jpg  \n",
            "  inflating: CNNDataset/Truck/9b6015d8a8eef681.jpg  \n",
            "  inflating: CNNDataset/Truck/41c86c5086a364ee.jpg  \n",
            "  inflating: CNNDataset/Truck/d43de8bea6942cea.jpg  \n",
            "  inflating: CNNDataset/Truck/d82f0dbc0628bd0a.jpg  \n",
            "  inflating: CNNDataset/Truck/e1f2f88f3ceffe12.jpg  \n",
            "  inflating: CNNDataset/Truck/c5a01e5aa60a278d.jpg  \n",
            "  inflating: CNNDataset/Truck/18e9888614878dc5.jpg  \n",
            "  inflating: CNNDataset/Truck/0edf74fc4b86603c.jpg  \n",
            "  inflating: CNNDataset/Truck/ee0ca6c6d52d9718.jpg  \n",
            "  inflating: CNNDataset/Truck/f0873ff6437f062e.jpg  \n",
            "  inflating: CNNDataset/Truck/b8e46f9cb1479a05.jpg  \n",
            "  inflating: CNNDataset/Truck/ea1fe32c118a045a.jpg  \n",
            "  inflating: CNNDataset/Truck/0194c725c0b9d072.jpg  \n",
            "  inflating: CNNDataset/Truck/0a9c6457ed1c64ff.jpg  \n",
            "  inflating: CNNDataset/Truck/5162b02f2a683ddc.jpg  \n",
            "  inflating: CNNDataset/Truck/82c4db39084116c9.jpg  \n",
            "  inflating: CNNDataset/Truck/75a2c38819ce82dc.jpg  \n",
            "  inflating: CNNDataset/Truck/64508a438525b1c3.jpg  \n",
            "  inflating: CNNDataset/Truck/1069bae36630a3b9.jpg  \n",
            "  inflating: CNNDataset/Truck/5513307d24b1ee04.jpg  \n",
            "  inflating: CNNDataset/Truck/723ce4c0651aa077.jpg  \n",
            "  inflating: CNNDataset/Truck/aafcd8efb9b331f8.jpg  \n",
            "  inflating: CNNDataset/Truck/bff6530aed9b61d0.jpg  \n",
            "  inflating: CNNDataset/Truck/ae240932dff9c305.jpg  \n",
            "  inflating: CNNDataset/Truck/5ba23dee5d1c4a8d.jpg  \n",
            "  inflating: CNNDataset/Truck/602ed24bda8d48a0.jpg  \n",
            "  inflating: CNNDataset/Truck/188c7894ee14d8a2.jpg  \n",
            "  inflating: CNNDataset/Truck/f1097181c29d9856.jpg  \n",
            "  inflating: CNNDataset/Truck/21c26e2577d503aa.jpg  \n",
            "  inflating: CNNDataset/Truck/6143fcc1dffe8ec0.jpg  \n",
            "  inflating: CNNDataset/Truck/902814138cb8b88c.jpg  \n",
            "  inflating: CNNDataset/Truck/9ca66bcd7a567008.jpg  \n",
            "  inflating: CNNDataset/Truck/3453a75611009a15.jpg  \n",
            "  inflating: CNNDataset/Truck/5d54c632f2afcbf0.jpg  \n",
            "  inflating: CNNDataset/Truck/c43a1e82d228434a.jpg  \n",
            "  inflating: CNNDataset/Truck/67361a6cb0d7720e.jpg  \n",
            "  inflating: CNNDataset/Truck/4857ee7806fceb42.jpg  \n",
            "  inflating: CNNDataset/Truck/463e18367335aa26.jpg  \n",
            "  inflating: CNNDataset/Truck/6484bc664019fff5.jpg  \n",
            "  inflating: CNNDataset/Truck/27094f06ed3f7408.jpg  \n",
            "  inflating: CNNDataset/Truck/9104937a1e47df43.jpg  \n",
            "  inflating: CNNDataset/Truck/6f1a94310d4b735e.jpg  \n",
            "  inflating: CNNDataset/Truck/f74d382757c8aa08.jpg  \n",
            "  inflating: CNNDataset/Truck/0077d127ba76adfb.jpg  \n",
            "  inflating: CNNDataset/Truck/7dfbb88ce96a3475.jpg  \n",
            "  inflating: CNNDataset/Truck/857677b62cdf4f49.jpg  \n",
            "  inflating: CNNDataset/Truck/00272dd158c0663a.jpg  \n",
            "  inflating: CNNDataset/Truck/f2fff5de2c15594f.jpg  \n",
            "  inflating: CNNDataset/Truck/0082c50527ec2640.jpg  \n",
            "  inflating: CNNDataset/Truck/32cc48b60184ec89.jpg  \n",
            "  inflating: CNNDataset/Truck/d84df20c677bcaaa.jpg  \n",
            "  inflating: CNNDataset/Truck/f6602cd8513969ca.jpg  \n",
            "  inflating: CNNDataset/Truck/3846ab7971da439c.jpg  \n",
            "  inflating: CNNDataset/Truck/c5790cbf5acd678e.jpg  \n",
            "  inflating: CNNDataset/Truck/8266a4f69172bb32.jpg  \n",
            "  inflating: CNNDataset/Truck/a7f307de5fcb6cb7.jpg  \n",
            "  inflating: CNNDataset/Truck/be922b2556072b88.jpg  \n",
            "  inflating: CNNDataset/Truck/205a2a181c2a6577.jpg  \n",
            "  inflating: CNNDataset/Truck/306826e693c6ecec.jpg  \n",
            "  inflating: CNNDataset/Truck/e26347d2e35079fd.jpg  \n",
            "  inflating: CNNDataset/Truck/850267ffbd1708d1.jpg  \n",
            "  inflating: CNNDataset/Truck/92ef84f2d0c3f3c0.jpg  \n",
            "  inflating: CNNDataset/Truck/a0755dae5354f121.jpg  \n",
            "  inflating: CNNDataset/Truck/d334d8ffe5acecb0.jpg  \n",
            "  inflating: CNNDataset/Truck/59612db58c2c984c.jpg  \n",
            "  inflating: CNNDataset/Truck/0034bd453281c508.jpg  \n",
            "  inflating: CNNDataset/Truck/25b25315ba8efcbe.jpg  \n",
            "  inflating: CNNDataset/Truck/a0ec021af5e23cdc.jpg  \n",
            "  inflating: CNNDataset/Truck/9ec347075ae3fdd1.jpg  \n",
            "  inflating: CNNDataset/Truck/813621aa6bc9415e.jpg  \n",
            "  inflating: CNNDataset/Truck/d4c8892b21d7e81d.jpg  \n",
            "  inflating: CNNDataset/Truck/0c092957ed736467.jpg  \n",
            "  inflating: CNNDataset/Truck/d6c81ef1470b6da3.jpg  \n",
            "  inflating: CNNDataset/Truck/e538b618fee24688.jpg  \n",
            "  inflating: CNNDataset/Truck/f2419884c07d0ad5.jpg  \n",
            "  inflating: CNNDataset/Truck/7d03d2c2ae766d8c.jpg  \n",
            "  inflating: CNNDataset/Truck/f844fd012d3d98c6.jpg  \n",
            "  inflating: CNNDataset/Truck/698be8f95c4f0fa5.jpg  \n",
            "  inflating: CNNDataset/Truck/9a2d057dee70b3f8.jpg  \n",
            "  inflating: CNNDataset/Truck/1c1860c2474b0261.jpg  \n",
            "  inflating: CNNDataset/Truck/328e7bede2cc3e5a.jpg  \n",
            "  inflating: CNNDataset/Truck/1f15be894373d45b.jpg  \n",
            "  inflating: CNNDataset/Truck/01644c049ceff0cf.jpg  \n",
            "  inflating: CNNDataset/Truck/0043da1ff6622b4f.jpg  \n",
            "  inflating: CNNDataset/Truck/1511d66b78d47041.jpg  \n",
            "  inflating: CNNDataset/Truck/84148087e14faaee.jpg  \n",
            "  inflating: CNNDataset/Truck/6e31ea5888f9b7ce.jpg  \n",
            "  inflating: CNNDataset/Truck/2bc7bd50787a052d.jpg  \n",
            "  inflating: CNNDataset/Truck/d5031b725a38af81.jpg  \n",
            "  inflating: CNNDataset/Truck/57d0447bf6ef13fd.jpg  \n",
            "  inflating: CNNDataset/Truck/edf74da11c0d1e6f.jpg  \n",
            "  inflating: CNNDataset/Truck/22d7dbe938240b58.jpg  \n",
            "  inflating: CNNDataset/Truck/dd934e47cdbd13de.jpg  \n",
            "  inflating: CNNDataset/Truck/43056f3048e80efd.jpg  \n",
            "  inflating: CNNDataset/Truck/4b5b2240698a743a.jpg  \n",
            "  inflating: CNNDataset/Truck/d929f6988ed08bbd.jpg  \n",
            "  inflating: CNNDataset/Truck/193de4a2b77a4fc4.jpg  \n",
            "  inflating: CNNDataset/Truck/1324de9e43e5ab16.jpg  \n",
            "  inflating: CNNDataset/Truck/7e5d1e7fcb44eedb.jpg  \n",
            "  inflating: CNNDataset/Truck/297c96f416bd3a13.jpg  \n",
            "  inflating: CNNDataset/Truck/0208c1440b1c56b5.jpg  \n",
            "  inflating: CNNDataset/Truck/9e4f6f06af74b574.jpg  \n",
            "  inflating: CNNDataset/Truck/4be14d014be16f0e.jpg  \n",
            "  inflating: CNNDataset/Truck/f7476d536a81a069.jpg  \n",
            "  inflating: CNNDataset/Truck/8b278a058f2de719.jpg  \n",
            "  inflating: CNNDataset/Truck/db6735de920bd862.jpg  \n",
            "  inflating: CNNDataset/Truck/7a04b7bc009c216f.jpg  \n",
            "  inflating: CNNDataset/Truck/543f2f502cb67191.jpg  \n",
            "  inflating: CNNDataset/Truck/36837ac3bfb46990.jpg  \n",
            "  inflating: CNNDataset/Truck/49304cbc80ffb423.jpg  \n",
            "  inflating: CNNDataset/Truck/16b723c453b3fa8d.jpg  \n",
            "  inflating: CNNDataset/Truck/df90d261113bcc49.jpg  \n",
            "  inflating: CNNDataset/Truck/7b2001f9447e235d.jpg  \n",
            "  inflating: CNNDataset/Truck/4d992b010e9fd12e.jpg  \n",
            "  inflating: CNNDataset/Truck/09feecbc4f018b2c.jpg  \n",
            "  inflating: CNNDataset/Truck/57f6cdd9057141ce.jpg  \n",
            "  inflating: CNNDataset/Truck/9d68892592167700.jpg  \n",
            "  inflating: CNNDataset/Truck/b3c8e839f1d7d9ce.jpg  \n",
            "  inflating: CNNDataset/Truck/0912661b887e7f11.jpg  \n",
            "  inflating: CNNDataset/Truck/1798c2b87f1fc312.jpg  \n",
            "  inflating: CNNDataset/Truck/bca9a3fea9d58f48.jpg  \n",
            "  inflating: CNNDataset/Truck/aa073232b89cd06d.jpg  \n",
            "  inflating: CNNDataset/Truck/6c6141157874e32d.jpg  \n",
            "  inflating: CNNDataset/Truck/dcd20711f1e04fba.jpg  \n",
            "  inflating: CNNDataset/Truck/fd7fcb7f01eea1c8.jpg  \n",
            "  inflating: CNNDataset/Truck/c70d62bffe00f159.jpg  \n",
            "  inflating: CNNDataset/Truck/5f16be94c7e701f6.jpg  \n",
            "  inflating: CNNDataset/Truck/8a5fdd0eb74ef538.jpg  \n",
            "  inflating: CNNDataset/Truck/1b33e9c90db5ca6a.jpg  \n",
            "  inflating: CNNDataset/Truck/dec8114781963e78.jpg  \n",
            "  inflating: CNNDataset/Truck/f0acec67f7c55695.jpg  \n",
            "  inflating: CNNDataset/Truck/682ae88fc292bba8.jpg  \n",
            "  inflating: CNNDataset/Truck/9d2cbe35a650b6b6.jpg  \n",
            "  inflating: CNNDataset/Truck/30737cb442acfd0e.jpg  \n",
            "  inflating: CNNDataset/Truck/68778b3e66bd53d2.jpg  \n",
            "  inflating: CNNDataset/Truck/6d29ebc9b9d2eed9.jpg  \n",
            "  inflating: CNNDataset/Truck/57eed39c48c01fbc.jpg  \n",
            "  inflating: CNNDataset/Truck/ff937e6c07042ea1.jpg  \n",
            "  inflating: CNNDataset/Truck/c4d05cc4ebb07dc4.jpg  \n",
            "  inflating: CNNDataset/Truck/72717e0911491997.jpg  \n",
            "  inflating: CNNDataset/Truck/3c67fbcdcda8ffd4.jpg  \n",
            "  inflating: CNNDataset/Truck/593510ef13551bf5.jpg  \n",
            "  inflating: CNNDataset/Truck/5b3d76f1c03289c6.jpg  \n",
            "  inflating: CNNDataset/Truck/66f6e6793898737c.jpg  \n",
            "  inflating: CNNDataset/Truck/ad0abd807a17905c.jpg  \n",
            "  inflating: CNNDataset/Truck/a1d8904fc1c86370.jpg  \n",
            "  inflating: CNNDataset/Truck/b9fd46b2e4ff4952.jpg  \n",
            "  inflating: CNNDataset/Truck/08d817d970f118f6.jpg  \n",
            "  inflating: CNNDataset/Truck/01e9c3d5f40b74fc.jpg  \n",
            "  inflating: CNNDataset/Truck/cda86b0e90e9d38f.jpg  \n",
            "  inflating: CNNDataset/Truck/d42973cf26871b9e.jpg  \n",
            "  inflating: CNNDataset/Truck/2809dbd6d10c3f46.jpg  \n",
            "  inflating: CNNDataset/Truck/b38072b0b237c9d1.jpg  \n",
            "  inflating: CNNDataset/Truck/2b13b42ee2689c83.jpg  \n",
            "  inflating: CNNDataset/Truck/3751b34d97d324c7.jpg  \n",
            "  inflating: CNNDataset/Truck/a2c40642a8f241bd.jpg  \n",
            "   creating: CNNDataset/Car/\n",
            "  inflating: CNNDataset/Car/a7287106de84b405.jpg  \n",
            "  inflating: CNNDataset/Car/67ff929ee77588f5.jpg  \n",
            "  inflating: CNNDataset/Car/9ed68b446821ac19.jpg  \n",
            "  inflating: CNNDataset/Car/44e69ad1abca4879.jpg  \n",
            "  inflating: CNNDataset/Car/28897a549dd42ffd.jpg  \n",
            "  inflating: CNNDataset/Car/d37871afe1db2189.jpg  \n",
            "  inflating: CNNDataset/Car/f1fb49b2de002de7.jpg  \n",
            "  inflating: CNNDataset/Car/cd2b9cf5e93b2c02.jpg  \n",
            "  inflating: CNNDataset/Car/af3db197b2f78d10.jpg  \n",
            "  inflating: CNNDataset/Car/b17cf83e563436c4.jpg  \n",
            "  inflating: CNNDataset/Car/2d94e4aa09c1c7de.jpg  \n",
            "  inflating: CNNDataset/Car/4c81bcf0f5c0e195.jpg  \n",
            "  inflating: CNNDataset/Car/ca3f47b64f45d627.jpg  \n",
            "  inflating: CNNDataset/Car/25c18ddccf314220.jpg  \n",
            "  inflating: CNNDataset/Car/63d8cc4af71f740b.jpg  \n",
            "  inflating: CNNDataset/Car/b28de23c0556f510.jpg  \n",
            "  inflating: CNNDataset/Car/b8eb956b942348f0.jpg  \n",
            "  inflating: CNNDataset/Car/16f68d119ee4cb6e.jpg  \n",
            "  inflating: CNNDataset/Car/985cd6e06088350a.jpg  \n",
            "  inflating: CNNDataset/Car/16a7b57fab951225.jpg  \n",
            "  inflating: CNNDataset/Car/34b68fd9baa24404.jpg  \n",
            "  inflating: CNNDataset/Car/80a572df1735f1f1.jpg  \n",
            "  inflating: CNNDataset/Car/1409cd2847078968.jpg  \n",
            "  inflating: CNNDataset/Car/67649826d2e490bc.jpg  \n",
            "  inflating: CNNDataset/Car/88e8489a5f8c9808.jpg  \n",
            "  inflating: CNNDataset/Car/6b9ad9f396d3ab3b.jpg  \n",
            "  inflating: CNNDataset/Car/1fc3e6254d880bf0.jpg  \n",
            "  inflating: CNNDataset/Car/366eedd9184ace28.jpg  \n",
            "  inflating: CNNDataset/Car/aa45667e2befafe6.jpg  \n",
            "  inflating: CNNDataset/Car/887894c12c624d51.jpg  \n",
            "  inflating: CNNDataset/Car/5bd555ff32badb8e.jpg  \n",
            "  inflating: CNNDataset/Car/632847529fb689b1.jpg  \n",
            "  inflating: CNNDataset/Car/5f43199d822b83ce.jpg  \n",
            "  inflating: CNNDataset/Car/5eea8a87336506d4.jpg  \n",
            "  inflating: CNNDataset/Car/5b3e5544a69c1cfd.jpg  \n",
            "  inflating: CNNDataset/Car/9b269929471405df.jpg  \n",
            "  inflating: CNNDataset/Car/8b8de7999ba14565.jpg  \n",
            "  inflating: CNNDataset/Car/20e0ba8e739845e4.jpg  \n",
            "  inflating: CNNDataset/Car/c70d9a21f454c152.jpg  \n",
            "  inflating: CNNDataset/Car/faf68d3facfd5c76.jpg  \n",
            "  inflating: CNNDataset/Car/99fd1c2a486a4886.jpg  \n",
            "  inflating: CNNDataset/Car/47c5e6164cdd9b21.jpg  \n",
            "  inflating: CNNDataset/Car/3bf31faa28637ff1.jpg  \n",
            "  inflating: CNNDataset/Car/bebc5a513a891bb8.jpg  \n",
            "  inflating: CNNDataset/Car/6dcaa5dffc600ba7.jpg  \n",
            "  inflating: CNNDataset/Car/bfb895005946177b.jpg  \n",
            "  inflating: CNNDataset/Car/ca1e52ea3d7481c6.jpg  \n",
            "  inflating: CNNDataset/Car/79c0076fe1465c5b.jpg  \n",
            "  inflating: CNNDataset/Car/18cd9d6586055bfb.jpg  \n",
            "  inflating: CNNDataset/Car/3081ec3d757d1159.jpg  \n",
            "  inflating: CNNDataset/Car/c0211dac41c27a1f.jpg  \n",
            "  inflating: CNNDataset/Car/f2d74d890a0a90ea.jpg  \n",
            "  inflating: CNNDataset/Car/e982746cb469b34d.jpg  \n",
            "  inflating: CNNDataset/Car/24789ce3fb99dd01.jpg  \n",
            "  inflating: CNNDataset/Car/354a049deb1fec5c.jpg  \n",
            "  inflating: CNNDataset/Car/1ca9f561d7693ac7.jpg  \n",
            "  inflating: CNNDataset/Car/0f2585efba395266.jpg  \n",
            "  inflating: CNNDataset/Car/6b935eeea6290736.jpg  \n",
            "  inflating: CNNDataset/Car/f72c6d39b8572939.jpg  \n",
            "  inflating: CNNDataset/Car/a31a700c0c8c421a.jpg  \n",
            "  inflating: CNNDataset/Car/07cca219d23fa8c2.jpg  \n",
            "  inflating: CNNDataset/Car/439fd803ec2d5477.jpg  \n",
            "  inflating: CNNDataset/Car/8098506635323131.jpg  \n",
            "  inflating: CNNDataset/Car/ec8485860f8279d1.jpg  \n",
            "  inflating: CNNDataset/Car/9ff9f4e5a74b01e1.jpg  \n",
            "  inflating: CNNDataset/Car/edf0ececa452b62b.jpg  \n",
            "  inflating: CNNDataset/Car/7d88f825dc209cc6.jpg  \n",
            "  inflating: CNNDataset/Car/f09e597c9b248652.jpg  \n",
            "  inflating: CNNDataset/Car/e340b149b6d3a1a8.jpg  \n",
            "  inflating: CNNDataset/Car/70f703b0f33504d7.jpg  \n",
            "  inflating: CNNDataset/Car/95b74af45cede0e5.jpg  \n",
            "  inflating: CNNDataset/Car/33977997b359ca17.jpg  \n",
            "  inflating: CNNDataset/Car/a08c3b9dda0564f6.jpg  \n",
            "  inflating: CNNDataset/Car/9af612e072ead16c.jpg  \n",
            "  inflating: CNNDataset/Car/a9d10206a54fa02a.jpg  \n",
            "  inflating: CNNDataset/Car/815f285de31aec30.jpg  \n",
            "  inflating: CNNDataset/Car/83e614bc6c51d886.jpg  \n",
            "  inflating: CNNDataset/Car/79b3153a2cbfbace.jpg  \n",
            "  inflating: CNNDataset/Car/030bdb29df1052c9.jpg  \n",
            "  inflating: CNNDataset/Car/7e7226d591ebfd86.jpg  \n",
            "  inflating: CNNDataset/Car/1906090e7380f4e2.jpg  \n",
            "  inflating: CNNDataset/Car/fd64c8292085e9d4.jpg  \n",
            "  inflating: CNNDataset/Car/70fa05543bfbc0cc.jpg  \n",
            "  inflating: CNNDataset/Car/23f16446ba410c0c.jpg  \n",
            "  inflating: CNNDataset/Car/b67259076bac7b17.jpg  \n",
            "  inflating: CNNDataset/Car/e61e3f85dfb48b13.jpg  \n",
            "  inflating: CNNDataset/Car/2e157494856727ec.jpg  \n",
            "  inflating: CNNDataset/Car/b3afc8866af1e8d1.jpg  \n",
            "  inflating: CNNDataset/Car/47e891e3eff4f748.jpg  \n",
            "  inflating: CNNDataset/Car/8e389979f586a493.jpg  \n",
            "  inflating: CNNDataset/Car/8df398ddc31e4b71.jpg  \n",
            "  inflating: CNNDataset/Car/02ab891ece5bd910.jpg  \n",
            "  inflating: CNNDataset/Car/1280626ac385df86.jpg  \n",
            "  inflating: CNNDataset/Car/a44cdf6c3945a657.jpg  \n",
            "  inflating: CNNDataset/Car/986dfd50f18f950c.jpg  \n",
            "  inflating: CNNDataset/Car/ae7f9c90b6c80027.jpg  \n",
            "  inflating: CNNDataset/Car/ef630da2f0deaf7d.jpg  \n",
            "  inflating: CNNDataset/Car/e9cc1045477c195d.jpg  \n",
            "  inflating: CNNDataset/Car/29118e479a10a9a2.jpg  \n",
            "  inflating: CNNDataset/Car/6d265d33c796e74d.jpg  \n",
            "  inflating: CNNDataset/Car/57ac99d1d5a3e8ca.jpg  \n",
            "  inflating: CNNDataset/Car/1736e07788834bd2.jpg  \n",
            "  inflating: CNNDataset/Car/fa2c651281a5f13a.jpg  \n",
            "  inflating: CNNDataset/Car/0311ec4be446d16a.jpg  \n",
            "  inflating: CNNDataset/Car/6aea2127c0859c68.jpg  \n",
            "  inflating: CNNDataset/Car/c468276392522126.jpg  \n",
            "  inflating: CNNDataset/Car/7526d94fa4d6f50d.jpg  \n",
            "  inflating: CNNDataset/Car/97673724106bccd8.jpg  \n",
            "  inflating: CNNDataset/Car/c06b991177edfd77.jpg  \n",
            "  inflating: CNNDataset/Car/2c8f2e7bf5d7f3ee.jpg  \n",
            "  inflating: CNNDataset/Car/c7ec7c8b5743c86d.jpg  \n",
            "  inflating: CNNDataset/Car/7cf55ac1e1553361.jpg  \n",
            "  inflating: CNNDataset/Car/f668125c27b06fb9.jpg  \n",
            "  inflating: CNNDataset/Car/d3275c3bc6619d75.jpg  \n",
            "  inflating: CNNDataset/Car/13a4b7c8ecdee07b.jpg  \n",
            "  inflating: CNNDataset/Car/cac00c2862fe580b.jpg  \n",
            "  inflating: CNNDataset/Car/4a9eb7f7f3dafbdc.jpg  \n",
            "  inflating: CNNDataset/Car/5d65868b539c676c.jpg  \n",
            "  inflating: CNNDataset/Car/d7bd9593219874bd.jpg  \n",
            "  inflating: CNNDataset/Car/93a59b0e9e7ea967.jpg  \n",
            "  inflating: CNNDataset/Car/c4b18e8801e139c4.jpg  \n",
            "  inflating: CNNDataset/Car/53efbbda29fd3651.jpg  \n",
            "  inflating: CNNDataset/Car/bdbd205e3ebb4ffb.jpg  \n",
            "  inflating: CNNDataset/Car/1f7620b0eb12f980.jpg  \n",
            "  inflating: CNNDataset/Car/f56b7cacc75c8aba.jpg  \n",
            "  inflating: CNNDataset/Car/627a3a195d5ad1fc.jpg  \n",
            "  inflating: CNNDataset/Car/d1c654bf04ceabd6.jpg  \n",
            "  inflating: CNNDataset/Car/19e4437f0c5f5c2a.jpg  \n",
            "  inflating: CNNDataset/Car/f673b8e39fc77fe6.jpg  \n",
            "  inflating: CNNDataset/Car/791c9361d08db1b2.jpg  \n",
            "  inflating: CNNDataset/Car/da1f86c39b52c084.jpg  \n",
            "  inflating: CNNDataset/Car/9128fa470ef13e21.jpg  \n",
            "  inflating: CNNDataset/Car/6d6c67b8ad66934b.jpg  \n",
            "  inflating: CNNDataset/Car/1b6c27a199efd739.jpg  \n",
            "  inflating: CNNDataset/Car/93c6f6ce788be728.jpg  \n",
            "  inflating: CNNDataset/Car/c6f31efdddd8bd12.jpg  \n",
            "  inflating: CNNDataset/Car/6fb697b222f334db.jpg  \n",
            "  inflating: CNNDataset/Car/841ddb01cab36cd4.jpg  \n",
            "  inflating: CNNDataset/Car/50a239ba3a99a85c.jpg  \n",
            "  inflating: CNNDataset/Car/a10033bd4442ad25.jpg  \n",
            "  inflating: CNNDataset/Car/46e4009df254bd70.jpg  \n",
            "  inflating: CNNDataset/Car/8e9ed7f23fa1c843.jpg  \n",
            "  inflating: CNNDataset/Car/f283e1d6b8fc9a21.jpg  \n",
            "  inflating: CNNDataset/Car/e93b9303bb126050.jpg  \n",
            "  inflating: CNNDataset/Car/1384b7b72f16c64f.jpg  \n",
            "  inflating: CNNDataset/Car/39006c92c9e86b86.jpg  \n",
            "  inflating: CNNDataset/Car/530141e148a3e2f8.jpg  \n",
            "  inflating: CNNDataset/Car/1b55621372668519.jpg  \n",
            "  inflating: CNNDataset/Car/a731395352f2459e.jpg  \n",
            "  inflating: CNNDataset/Car/e2b7c9f1191d5058.jpg  \n",
            "  inflating: CNNDataset/Car/d9c7e178fa11afc8.jpg  \n",
            "  inflating: CNNDataset/Car/05063f267395efe0.jpg  \n",
            "  inflating: CNNDataset/Car/2c588ceee6dca5ad.jpg  \n",
            "  inflating: CNNDataset/Car/abf6e2c1ee157881.jpg  \n",
            "  inflating: CNNDataset/Car/45b681c0c8a41f8a.jpg  \n",
            "  inflating: CNNDataset/Car/202bc4ed1ad81f8b.jpg  \n",
            "  inflating: CNNDataset/Car/d25d76df61bb1d73.jpg  \n",
            "  inflating: CNNDataset/Car/4eb7f30136d23c87.jpg  \n",
            "  inflating: CNNDataset/Car/a7d00c8818bce116.jpg  \n",
            "  inflating: CNNDataset/Car/5c00f9c4e40e3121.jpg  \n",
            "  inflating: CNNDataset/Car/06f873f29b069268.jpg  \n",
            "  inflating: CNNDataset/Car/a38d17ea37f0a9c0.jpg  \n",
            "  inflating: CNNDataset/Car/6deb277cd838fdec.jpg  \n",
            "  inflating: CNNDataset/Car/9646bf2d13eb01e7.jpg  \n",
            "  inflating: CNNDataset/Car/a75a00bd73da7d2d.jpg  \n",
            "  inflating: CNNDataset/Car/cc1bee9a818572bc.jpg  \n",
            "  inflating: CNNDataset/Car/80d97c14cafcf987.jpg  \n",
            "  inflating: CNNDataset/Car/0eeb7b2354e16d73.jpg  \n",
            "  inflating: CNNDataset/Car/17819423a12fa572.jpg  \n",
            "  inflating: CNNDataset/Car/ed3f3c4eddba8115.jpg  \n",
            "  inflating: CNNDataset/Car/9c1d016b7d327f16.jpg  \n",
            "  inflating: CNNDataset/Car/ae23406db65263fe.jpg  \n",
            "  inflating: CNNDataset/Car/a5d95cd85e30518f.jpg  \n",
            "  inflating: CNNDataset/Car/d13b098523bb49f2.jpg  \n",
            "  inflating: CNNDataset/Car/c5bbef7d1819fccd.jpg  \n",
            "  inflating: CNNDataset/Car/0826a7c8746b1625.jpg  \n",
            "  inflating: CNNDataset/Car/9ec05f81db79c58e.jpg  \n",
            "  inflating: CNNDataset/Car/016facd97be00ec9.jpg  \n",
            "  inflating: CNNDataset/Car/8e8b56a361505be1.jpg  \n",
            "  inflating: CNNDataset/Car/0e06ee2e62ffb1a4.jpg  \n",
            "  inflating: CNNDataset/Car/8dd3becc83907e3b.jpg  \n",
            "  inflating: CNNDataset/Car/d0d72976a4ef259b.jpg  \n",
            "  inflating: CNNDataset/Car/f0bb852b0af58df8.jpg  \n",
            "  inflating: CNNDataset/Car/0025b4435a0ac672.jpg  \n",
            "  inflating: CNNDataset/Car/d13dae37b82381b8.jpg  \n",
            "  inflating: CNNDataset/Car/21e0af3a96cdb9fa.jpg  \n",
            "  inflating: CNNDataset/Car/6249edfc0d65a59d.jpg  \n",
            "   creating: CNNDataset/train/\n",
            "  inflating: CNNDataset/train/7cc6e09f8f1d2698.jpg  \n",
            "  inflating: CNNDataset/train/64508a438525b1c3.jpg  \n",
            "  inflating: CNNDataset/train/18e9888614878dc5.jpg  \n",
            "  inflating: CNNDataset/train/c70d62bffe00f159.jpg  \n",
            "  inflating: CNNDataset/train/08d817d970f118f6.jpg  \n",
            "  inflating: CNNDataset/train/698be8f95c4f0fa5.jpg  \n",
            "  inflating: CNNDataset/train/dd934e47cdbd13de.jpg  \n",
            "  inflating: CNNDataset/train/0194c725c0b9d072.jpg  \n",
            "  inflating: CNNDataset/train/7a04b7bc009c216f.jpg  \n",
            "  inflating: CNNDataset/train/72717e0911491997.jpg  \n",
            "  inflating: CNNDataset/train/9104937a1e47df43.jpg  \n",
            "  inflating: CNNDataset/train/21c26e2577d503aa.jpg  \n",
            "  inflating: CNNDataset/train/01644c049ceff0cf.jpg  \n",
            "  inflating: CNNDataset/train/9e4f6f06af74b574.jpg  \n",
            "  inflating: CNNDataset/train/c5790cbf5acd678e.jpg  \n",
            "  inflating: CNNDataset/train/84148087e14faaee.jpg  \n",
            "  inflating: CNNDataset/train/9d2cbe35a650b6b6.jpg  \n",
            "  inflating: CNNDataset/train/593510ef13551bf5.jpg  \n",
            "  inflating: CNNDataset/train/7dfbb88ce96a3475.jpg  \n",
            "  inflating: CNNDataset/train/e1dc0a2765038997.jpg  \n",
            "  inflating: CNNDataset/train/0082c50527ec2640.jpg  \n",
            "  inflating: CNNDataset/train/606e8d152ae2c704.jpg  \n",
            "  inflating: CNNDataset/train/21f283bece63c7c9.jpg  \n",
            "  inflating: CNNDataset/train/5b3d76f1c03289c6.jpg  \n",
            "  inflating: CNNDataset/train/cda86b0e90e9d38f.jpg  \n",
            "  inflating: CNNDataset/train/b9fd46b2e4ff4952.jpg  \n",
            "  inflating: CNNDataset/train/d84df20c677bcaaa.jpg  \n",
            "  inflating: CNNDataset/train/5ba23dee5d1c4a8d.jpg  \n",
            "  inflating: CNNDataset/train/1b33e9c90db5ca6a.jpg  \n",
            "  inflating: CNNDataset/train/57f6cdd9057141ce.jpg  \n",
            "  inflating: CNNDataset/train/f5ff3013e4f0b1d1.jpg  \n",
            "  inflating: CNNDataset/train/602ed24bda8d48a0.jpg  \n",
            "  inflating: CNNDataset/train/d82f0dbc0628bd0a.jpg  \n",
            "  inflating: CNNDataset/train/e538b618fee24688.jpg  \n",
            "  inflating: CNNDataset/train/2809dbd6d10c3f46.jpg  \n",
            "  inflating: CNNDataset/train/682ae88fc292bba8.jpg  \n",
            "  inflating: CNNDataset/train/ad0abd807a17905c.jpg  \n",
            "  inflating: CNNDataset/train/a2c40642a8f241bd.jpg  \n",
            "  inflating: CNNDataset/train/57eed39c48c01fbc.jpg  \n",
            "  inflating: CNNDataset/train/b3c8e839f1d7d9ce.jpg  \n",
            "  inflating: CNNDataset/train/bca9a3fea9d58f48.jpg  \n",
            "  inflating: CNNDataset/train/b8e46f9cb1479a05.jpg  \n",
            "  inflating: CNNDataset/train/3846ab7971da439c.jpg  \n",
            "  inflating: CNNDataset/train/857677b62cdf4f49.jpg  \n",
            "  inflating: CNNDataset/train/30737cb442acfd0e.jpg  \n",
            "  inflating: CNNDataset/train/188c7894ee14d8a2.jpg  \n",
            "  inflating: CNNDataset/train/0208c1440b1c56b5.jpg  \n",
            "  inflating: CNNDataset/train/328e7bede2cc3e5a.jpg  \n",
            "  inflating: CNNDataset/train/0b8f012362e805c9.jpg  \n",
            "  inflating: CNNDataset/train/6e31ea5888f9b7ce.jpg  \n",
            "  inflating: CNNDataset/train/a709f87623a9e4b9.jpg  \n",
            "  inflating: CNNDataset/train/0786212bde3d2c9c.jpg  \n",
            "  inflating: CNNDataset/train/32cc48b60184ec89.jpg  \n",
            "  inflating: CNNDataset/train/68778b3e66bd53d2.jpg  \n",
            "  inflating: CNNDataset/train/85151615a3dd7394.jpg  \n",
            "  inflating: CNNDataset/train/205a2a181c2a6577.jpg  \n",
            "  inflating: CNNDataset/train/0912661b887e7f11.jpg  \n",
            "  inflating: CNNDataset/train/bff6530aed9b61d0.jpg  \n",
            "  inflating: CNNDataset/train/faf8854c355a752a.jpg  \n",
            "  inflating: CNNDataset/train/3751b34d97d324c7.jpg  \n",
            "  inflating: CNNDataset/train/193de4a2b77a4fc4.jpg  \n",
            "  inflating: CNNDataset/train/e26347d2e35079fd.jpg  \n",
            "  inflating: CNNDataset/train/55d9de60ad6279c5.jpg  \n",
            "  inflating: CNNDataset/train/7d03d2c2ae766d8c.jpg  \n",
            "  inflating: CNNDataset/train/36837ac3bfb46990.jpg  \n",
            "  inflating: CNNDataset/train/f7476d536a81a069.jpg  \n",
            "  inflating: CNNDataset/train/e8a2a51e75297966.jpg  \n",
            "  inflating: CNNDataset/train/543f2f502cb67191.jpg  \n",
            "  inflating: CNNDataset/train/f74d382757c8aa08.jpg  \n",
            "  inflating: CNNDataset/train/dac8ac5ad2bca314.jpg  \n",
            "  inflating: CNNDataset/train/463e18367335aa26.jpg  \n",
            "  inflating: CNNDataset/train/41c86c5086a364ee.jpg  \n",
            "  inflating: CNNDataset/train/aafcd8efb9b331f8.jpg  \n",
            "  inflating: CNNDataset/train/6143fcc1dffe8ec0.jpg  \n",
            "  inflating: CNNDataset/train/b1a122749844023d.jpg  \n",
            "  inflating: CNNDataset/train/e1f2f88f3ceffe12.jpg  \n",
            "  inflating: CNNDataset/train/6c6141157874e32d.jpg  \n",
            "  inflating: CNNDataset/train/297c96f416bd3a13.jpg  \n",
            "  inflating: CNNDataset/train/f2fff5de2c15594f.jpg  \n",
            "  inflating: CNNDataset/train/c4d05cc4ebb07dc4.jpg  \n",
            "  inflating: CNNDataset/train/67361a6cb0d7720e.jpg  \n",
            "  inflating: CNNDataset/train/df6cfb5e5f9f967d.jpg  \n",
            "  inflating: CNNDataset/train/7b2001f9447e235d.jpg  \n",
            "  inflating: CNNDataset/train/df90d261113bcc49.jpg  \n",
            "  inflating: CNNDataset/train/71cc934013f02bfe.jpg  \n",
            "  inflating: CNNDataset/train/0c092957ed736467.jpg  \n",
            "  inflating: CNNDataset/train/1f15be894373d45b.jpg  \n",
            "  inflating: CNNDataset/train/3453a75611009a15.jpg  \n",
            "  inflating: CNNDataset/train/92ef84f2d0c3f3c0.jpg  \n",
            "  inflating: CNNDataset/train/a7f307de5fcb6cb7.jpg  \n",
            "  inflating: CNNDataset/train/d6c81ef1470b6da3.jpg  \n",
            "  inflating: CNNDataset/train/22d7dbe938240b58.jpg  \n",
            "  inflating: CNNDataset/train/b82b8ddb96dd6f3a.jpg  \n",
            "  inflating: CNNDataset/train/a1d8904fc1c86370.jpg  \n",
            "  inflating: CNNDataset/train/1c1860c2474b0261.jpg  \n",
            "  inflating: CNNDataset/train/ff937e6c07042ea1.jpg  \n",
            "  inflating: CNNDataset/train/edf74da11c0d1e6f.jpg  \n",
            "  inflating: CNNDataset/train/49304cbc80ffb423.jpg  \n",
            "  inflating: CNNDataset/train/5162b02f2a683ddc.jpg  \n",
            "  inflating: CNNDataset/train/9b6015d8a8eef681.jpg  \n",
            "  inflating: CNNDataset/train/dcd20711f1e04fba.jpg  \n",
            "  inflating: CNNDataset/train/582df91290740cfa.jpg  \n",
            "  inflating: CNNDataset/train/9ca66bcd7a567008.jpg  \n",
            "  inflating: CNNDataset/train/0a5b2310fe6e429f.jpg  \n",
            "  inflating: CNNDataset/train/813621aa6bc9415e.jpg  \n",
            "  inflating: CNNDataset/train/5513307d24b1ee04.jpg  \n",
            "  inflating: CNNDataset/train/0034bd453281c508.jpg  \n",
            "  inflating: CNNDataset/train/4b5b2240698a743a.jpg  \n",
            "  inflating: CNNDataset/train/8b278a058f2de719.jpg  \n",
            "  inflating: CNNDataset/train/16b723c453b3fa8d.jpg  \n",
            "  inflating: CNNDataset/train/d334d8ffe5acecb0.jpg  \n",
            "  inflating: CNNDataset/train/d43de8bea6942cea.jpg  \n",
            "  inflating: CNNDataset/train/25b25315ba8efcbe.jpg  \n",
            "  inflating: CNNDataset/train/b38072b0b237c9d1.jpg  \n",
            "  inflating: CNNDataset/train/d5031b725a38af81.jpg  \n",
            "  inflating: CNNDataset/train/0043da1ff6622b4f.jpg  \n",
            "  inflating: CNNDataset/train/a0ec021af5e23cdc.jpg  \n",
            "  inflating: CNNDataset/train/3c67fbcdcda8ffd4.jpg  \n",
            "  inflating: CNNDataset/train/306826e693c6ecec.jpg  \n",
            "  inflating: CNNDataset/train/75a2c38819ce82dc.jpg  \n",
            "  inflating: CNNDataset/train/4be14d014be16f0e.jpg  \n",
            "  inflating: CNNDataset/train/1798c2b87f1fc312.jpg  \n",
            "  inflating: CNNDataset/train/2bc7bd50787a052d.jpg  \n",
            "  inflating: CNNDataset/train/f6602cd8513969ca.jpg  \n",
            "  inflating: CNNDataset/train/c5a01e5aa60a278d.jpg  \n",
            "  inflating: CNNDataset/train/4857ee7806fceb42.jpg  \n",
            "  inflating: CNNDataset/train/db6735de920bd862.jpg  \n",
            "  inflating: CNNDataset/train/f844fd012d3d98c6.jpg  \n",
            "  inflating: CNNDataset/train/f0873ff6437f062e.jpg  \n",
            "  inflating: CNNDataset/train/8266a4f69172bb32.jpg  \n",
            "  inflating: CNNDataset/train/723ce4c0651aa077.jpg  \n",
            "  inflating: CNNDataset/train/9ec347075ae3fdd1.jpg  \n",
            "  inflating: CNNDataset/train/0a9c6457ed1c64ff.jpg  \n",
            "  inflating: CNNDataset/train/d4c8892b21d7e81d.jpg  \n",
            "  inflating: CNNDataset/train/9d68892592167700.jpg  \n",
            "  inflating: CNNDataset/train/66f6e6793898737c.jpg  \n",
            "  inflating: CNNDataset/train/01e9c3d5f40b74fc.jpg  \n",
            "  inflating: CNNDataset/train/09feecbc4f018b2c.jpg  \n",
            "  inflating: CNNDataset/train/ee0ca6c6d52d9718.jpg  \n",
            "  inflating: CNNDataset/train/f0acec67f7c55695.jpg  \n",
            "  inflating: CNNDataset/train/6f1a94310d4b735e.jpg  \n",
            "  inflating: CNNDataset/train/82c4db39084116c9.jpg  \n",
            "  inflating: CNNDataset/train/9001036c66a19a83.jpg  \n",
            "  inflating: CNNDataset/train/6d29ebc9b9d2eed9.jpg  \n",
            "  inflating: CNNDataset/train/2b13b42ee2689c83.jpg  \n",
            "  inflating: CNNDataset/train/be922b2556072b88.jpg  \n",
            "  inflating: CNNDataset/train/6484bc664019fff5.jpg  \n",
            "  inflating: CNNDataset/train/4d992b010e9fd12e.jpg  \n",
            "  inflating: CNNDataset/train/5d54c632f2afcbf0.jpg  \n",
            "  inflating: CNNDataset/train/0edf74fc4b86603c.jpg  \n",
            "  inflating: CNNDataset/train/f1097181c29d9856.jpg  \n",
            "  inflating: CNNDataset/train/0077d127ba76adfb.jpg  \n",
            "  inflating: CNNDataset/train/57d0447bf6ef13fd.jpg  \n",
            "  inflating: CNNDataset/train/06ab38835a7a8a58.jpg  \n",
            "  inflating: CNNDataset/train/59612db58c2c984c.jpg  \n",
            "  inflating: CNNDataset/train/d42973cf26871b9e.jpg  \n",
            "  inflating: CNNDataset/train/ea1fe32c118a045a.jpg  \n",
            "  inflating: CNNDataset/train/27094f06ed3f7408.jpg  \n",
            "  inflating: CNNDataset/train/1511d66b78d47041.jpg  \n",
            "  inflating: CNNDataset/train/89eb16acf43df012.jpg  \n",
            "  inflating: CNNDataset/train/439fd803ec2d5477.jpg  \n",
            "  inflating: CNNDataset/train/93a59b0e9e7ea967.jpg  \n",
            "  inflating: CNNDataset/train/06f873f29b069268.jpg  \n",
            "  inflating: CNNDataset/train/f1fb49b2de002de7.jpg  \n",
            "  inflating: CNNDataset/train/cd2b9cf5e93b2c02.jpg  \n",
            "  inflating: CNNDataset/train/ed3f3c4eddba8115.jpg  \n",
            "  inflating: CNNDataset/train/97673724106bccd8.jpg  \n",
            "  inflating: CNNDataset/train/80d97c14cafcf987.jpg  \n",
            "  inflating: CNNDataset/train/2c588ceee6dca5ad.jpg  \n",
            "  inflating: CNNDataset/train/9128fa470ef13e21.jpg  \n",
            "  inflating: CNNDataset/train/ca1e52ea3d7481c6.jpg  \n",
            "  inflating: CNNDataset/train/986dfd50f18f950c.jpg  \n",
            "  inflating: CNNDataset/train/da1f86c39b52c084.jpg  \n",
            "  inflating: CNNDataset/train/5bd555ff32badb8e.jpg  \n",
            "  inflating: CNNDataset/train/ae7f9c90b6c80027.jpg  \n",
            "  inflating: CNNDataset/train/20e0ba8e739845e4.jpg  \n",
            "  inflating: CNNDataset/train/f673b8e39fc77fe6.jpg  \n",
            "  inflating: CNNDataset/train/4a9eb7f7f3dafbdc.jpg  \n",
            "  inflating: CNNDataset/train/7526d94fa4d6f50d.jpg  \n",
            "  inflating: CNNDataset/train/8e389979f586a493.jpg  \n",
            "  inflating: CNNDataset/train/8df398ddc31e4b71.jpg  \n",
            "  inflating: CNNDataset/train/a75a00bd73da7d2d.jpg  \n",
            "  inflating: CNNDataset/train/4eb7f30136d23c87.jpg  \n",
            "  inflating: CNNDataset/train/6d265d33c796e74d.jpg  \n",
            "  inflating: CNNDataset/train/70f703b0f33504d7.jpg  \n",
            "  inflating: CNNDataset/train/28897a549dd42ffd.jpg  \n",
            "  inflating: CNNDataset/train/5f43199d822b83ce.jpg  \n",
            "  inflating: CNNDataset/train/bdbd205e3ebb4ffb.jpg  \n",
            "  inflating: CNNDataset/train/79c0076fe1465c5b.jpg  \n",
            "  inflating: CNNDataset/train/d3275c3bc6619d75.jpg  \n",
            "  inflating: CNNDataset/train/edf0ececa452b62b.jpg  \n",
            "  inflating: CNNDataset/train/841ddb01cab36cd4.jpg  \n",
            "  inflating: CNNDataset/train/07cca219d23fa8c2.jpg  \n",
            "  inflating: CNNDataset/train/e2b7c9f1191d5058.jpg  \n",
            "  inflating: CNNDataset/train/99fd1c2a486a4886.jpg  \n",
            "  inflating: CNNDataset/train/ef630da2f0deaf7d.jpg  \n",
            "  inflating: CNNDataset/train/abf6e2c1ee157881.jpg  \n",
            "  inflating: CNNDataset/train/9c1d016b7d327f16.jpg  \n",
            "  inflating: CNNDataset/train/6249edfc0d65a59d.jpg  \n",
            "  inflating: CNNDataset/train/25c18ddccf314220.jpg  \n",
            "  inflating: CNNDataset/train/9646bf2d13eb01e7.jpg  \n",
            "  inflating: CNNDataset/train/bebc5a513a891bb8.jpg  \n",
            "  inflating: CNNDataset/train/1fc3e6254d880bf0.jpg  \n",
            "  inflating: CNNDataset/train/02ab891ece5bd910.jpg  \n",
            "  inflating: CNNDataset/train/a7d00c8818bce116.jpg  \n",
            "  inflating: CNNDataset/train/aa45667e2befafe6.jpg  \n",
            "  inflating: CNNDataset/train/a10033bd4442ad25.jpg  \n",
            "  inflating: CNNDataset/train/1ca9f561d7693ac7.jpg  \n",
            "  inflating: CNNDataset/train/6b9ad9f396d3ab3b.jpg  \n",
            "  inflating: CNNDataset/train/815f285de31aec30.jpg  \n",
            "  inflating: CNNDataset/train/f09e597c9b248652.jpg  \n",
            "  inflating: CNNDataset/train/a5d95cd85e30518f.jpg  \n",
            "  inflating: CNNDataset/train/a9d10206a54fa02a.jpg  \n",
            "  inflating: CNNDataset/train/d7bd9593219874bd.jpg  \n",
            "  inflating: CNNDataset/train/985cd6e06088350a.jpg  \n",
            "  inflating: CNNDataset/train/e340b149b6d3a1a8.jpg  \n",
            "  inflating: CNNDataset/train/a31a700c0c8c421a.jpg  \n",
            "  inflating: CNNDataset/train/83e614bc6c51d886.jpg  \n",
            "  inflating: CNNDataset/train/63d8cc4af71f740b.jpg  \n",
            "  inflating: CNNDataset/train/d13dae37b82381b8.jpg  \n",
            "  inflating: CNNDataset/train/29118e479a10a9a2.jpg  \n",
            "  inflating: CNNDataset/train/a08c3b9dda0564f6.jpg  \n",
            "  inflating: CNNDataset/train/627a3a195d5ad1fc.jpg  \n",
            "  inflating: CNNDataset/train/2d94e4aa09c1c7de.jpg  \n",
            "  inflating: CNNDataset/train/887894c12c624d51.jpg  \n",
            "  inflating: CNNDataset/train/9b269929471405df.jpg  \n",
            "  inflating: CNNDataset/train/9af612e072ead16c.jpg  \n",
            "  inflating: CNNDataset/train/a38d17ea37f0a9c0.jpg  \n",
            "  inflating: CNNDataset/train/d1c654bf04ceabd6.jpg  \n",
            "  inflating: CNNDataset/train/6deb277cd838fdec.jpg  \n",
            "  inflating: CNNDataset/train/57ac99d1d5a3e8ca.jpg  \n",
            "  inflating: CNNDataset/train/ec8485860f8279d1.jpg  \n",
            "  inflating: CNNDataset/train/ca3f47b64f45d627.jpg  \n",
            "  inflating: CNNDataset/train/ae23406db65263fe.jpg  \n",
            "  inflating: CNNDataset/train/0e06ee2e62ffb1a4.jpg  \n",
            "  inflating: CNNDataset/train/5c00f9c4e40e3121.jpg  \n",
            "  inflating: CNNDataset/train/95b74af45cede0e5.jpg  \n",
            "  inflating: CNNDataset/train/6fb697b222f334db.jpg  \n",
            "  inflating: CNNDataset/train/17819423a12fa572.jpg  \n",
            "  inflating: CNNDataset/train/5d65868b539c676c.jpg  \n",
            "  inflating: CNNDataset/train/d13b098523bb49f2.jpg  \n",
            "  inflating: CNNDataset/train/67ff929ee77588f5.jpg  \n",
            "  inflating: CNNDataset/train/fd64c8292085e9d4.jpg  \n",
            "  inflating: CNNDataset/train/24789ce3fb99dd01.jpg  \n",
            "  inflating: CNNDataset/train/8dd3becc83907e3b.jpg  \n",
            "  inflating: CNNDataset/train/fa2c651281a5f13a.jpg  \n",
            "  inflating: CNNDataset/train/2c8f2e7bf5d7f3ee.jpg  \n",
            "  inflating: CNNDataset/train/016facd97be00ec9.jpg  \n",
            "  inflating: CNNDataset/train/21e0af3a96cdb9fa.jpg  \n",
            "  inflating: CNNDataset/train/e93b9303bb126050.jpg  \n",
            "  inflating: CNNDataset/train/b67259076bac7b17.jpg  \n",
            "  inflating: CNNDataset/train/67649826d2e490bc.jpg  \n",
            "  inflating: CNNDataset/train/d25d76df61bb1d73.jpg  \n",
            "  inflating: CNNDataset/train/c4b18e8801e139c4.jpg  \n",
            "  inflating: CNNDataset/train/2e157494856727ec.jpg  \n",
            "  inflating: CNNDataset/train/47c5e6164cdd9b21.jpg  \n",
            "  inflating: CNNDataset/train/d37871afe1db2189.jpg  \n",
            "  inflating: CNNDataset/train/53efbbda29fd3651.jpg  \n",
            "  inflating: CNNDataset/train/6dcaa5dffc600ba7.jpg  \n",
            "  inflating: CNNDataset/train/23f16446ba410c0c.jpg  \n",
            "  inflating: CNNDataset/train/b8eb956b942348f0.jpg  \n",
            "  inflating: CNNDataset/train/13a4b7c8ecdee07b.jpg  \n",
            "  inflating: CNNDataset/train/030bdb29df1052c9.jpg  \n",
            "  inflating: CNNDataset/train/f0bb852b0af58df8.jpg  \n",
            "  inflating: CNNDataset/train/6d6c67b8ad66934b.jpg  \n",
            "  inflating: CNNDataset/train/e982746cb469b34d.jpg  \n",
            "  inflating: CNNDataset/train/e61e3f85dfb48b13.jpg  \n",
            "  inflating: CNNDataset/train/44e69ad1abca4879.jpg  \n",
            "  inflating: CNNDataset/train/c06b991177edfd77.jpg  \n",
            "  inflating: CNNDataset/train/c70d9a21f454c152.jpg  \n",
            "  inflating: CNNDataset/train/5b3e5544a69c1cfd.jpg  \n",
            "  inflating: CNNDataset/train/f56b7cacc75c8aba.jpg  \n",
            "  inflating: CNNDataset/train/0025b4435a0ac672.jpg  \n",
            "  inflating: CNNDataset/train/8e8b56a361505be1.jpg  \n",
            "  inflating: CNNDataset/train/9ff9f4e5a74b01e1.jpg  \n",
            "  inflating: CNNDataset/train/faf68d3facfd5c76.jpg  \n",
            "  inflating: CNNDataset/train/9ec05f81db79c58e.jpg  \n",
            "  inflating: CNNDataset/train/6aea2127c0859c68.jpg  \n",
            "  inflating: CNNDataset/train/80a572df1735f1f1.jpg  \n",
            "  inflating: CNNDataset/train/c0211dac41c27a1f.jpg  \n",
            "  inflating: CNNDataset/train/1f7620b0eb12f980.jpg  \n",
            "  inflating: CNNDataset/train/a44cdf6c3945a657.jpg  \n",
            "  inflating: CNNDataset/train/50a239ba3a99a85c.jpg  \n",
            "  inflating: CNNDataset/train/0826a7c8746b1625.jpg  \n",
            "  inflating: CNNDataset/train/47e891e3eff4f748.jpg  \n",
            "  inflating: CNNDataset/train/8b8de7999ba14565.jpg  \n",
            "  inflating: CNNDataset/train/0311ec4be446d16a.jpg  \n",
            "  inflating: CNNDataset/train/f668125c27b06fb9.jpg  \n",
            "  inflating: CNNDataset/train/c7ec7c8b5743c86d.jpg  \n",
            "  inflating: CNNDataset/train/cc1bee9a818572bc.jpg  \n",
            "  inflating: CNNDataset/train/cac00c2862fe580b.jpg  \n",
            "  inflating: CNNDataset/train/791c9361d08db1b2.jpg  \n",
            "  inflating: CNNDataset/train/c5bbef7d1819fccd.jpg  \n",
            "  inflating: CNNDataset/train/632847529fb689b1.jpg  \n",
            "  inflating: CNNDataset/train/bfb895005946177b.jpg  \n",
            "  inflating: CNNDataset/train/45b681c0c8a41f8a.jpg  \n",
            "  inflating: CNNDataset/train/1280626ac385df86.jpg  \n",
            "  inflating: CNNDataset/train/93c6f6ce788be728.jpg  \n",
            "  inflating: CNNDataset/train/f72c6d39b8572939.jpg  \n",
            "  inflating: CNNDataset/train/3bf31faa28637ff1.jpg  \n",
            "  inflating: CNNDataset/train/7d88f825dc209cc6.jpg  \n",
            "  inflating: CNNDataset/train/05063f267395efe0.jpg  \n",
            "  inflating: CNNDataset/train/16a7b57fab951225.jpg  \n",
            "  inflating: CNNDataset/train/8e9ed7f23fa1c843.jpg  \n",
            "  inflating: CNNDataset/train/c468276392522126.jpg  \n",
            "  inflating: CNNDataset/train/88e8489a5f8c9808.jpg  \n",
            "  inflating: CNNDataset/train/3081ec3d757d1159.jpg  \n",
            "  inflating: CNNDataset/train/1906090e7380f4e2.jpg  \n",
            "  inflating: CNNDataset/train/18cd9d6586055bfb.jpg  \n",
            "  inflating: CNNDataset/train/79b3153a2cbfbace.jpg  \n",
            "  inflating: CNNDataset/train/f283e1d6b8fc9a21.jpg  \n",
            "  inflating: CNNDataset/train/b17cf83e563436c4.jpg  \n",
            "  inflating: CNNDataset/train/39006c92c9e86b86.jpg  \n",
            "  inflating: CNNDataset/train/530141e148a3e2f8.jpg  \n",
            "  inflating: CNNDataset/train/70fa05543bfbc0cc.jpg  \n",
            "  inflating: CNNDataset/train/1b55621372668519.jpg  \n",
            "  inflating: CNNDataset/train/354a049deb1fec5c.jpg  \n",
            "  inflating: CNNDataset/train/7e7226d591ebfd86.jpg  \n",
            "  inflating: CNNDataset/train/33977997b359ca17.jpg  \n",
            "  inflating: CNNDataset/train/9ed68b446821ac19.jpg  \n",
            "   creating: CNNDataset/test/\n",
            "  inflating: CNNDataset/test/00e577cb0dd95bf4.jpg  \n",
            "  inflating: CNNDataset/test/fd7fcb7f01eea1c8.jpg  \n",
            "  inflating: CNNDataset/test/1069bae36630a3b9.jpg  \n",
            "  inflating: CNNDataset/test/9a2d057dee70b3f8.jpg  \n",
            "  inflating: CNNDataset/test/bff97f2b287f543f.jpg  \n",
            "  inflating: CNNDataset/test/5f16be94c7e701f6.jpg  \n",
            "  inflating: CNNDataset/test/ae240932dff9c305.jpg  \n",
            "  inflating: CNNDataset/test/a0755dae5354f121.jpg  \n",
            "  inflating: CNNDataset/test/1324de9e43e5ab16.jpg  \n",
            "  inflating: CNNDataset/test/00272dd158c0663a.jpg  \n",
            "  inflating: CNNDataset/test/ea87ffff22c14e7a.jpg  \n",
            "  inflating: CNNDataset/test/dec8114781963e78.jpg  \n",
            "  inflating: CNNDataset/test/d929f6988ed08bbd.jpg  \n",
            "  inflating: CNNDataset/test/43056f3048e80efd.jpg  \n",
            "  inflating: CNNDataset/test/850267ffbd1708d1.jpg  \n",
            "  inflating: CNNDataset/test/7e5d1e7fcb44eedb.jpg  \n",
            "  inflating: CNNDataset/test/c43a1e82d228434a.jpg  \n",
            "  inflating: CNNDataset/test/f2419884c07d0ad5.jpg  \n",
            "  inflating: CNNDataset/test/902814138cb8b88c.jpg  \n",
            "  inflating: CNNDataset/test/8a5fdd0eb74ef538.jpg  \n",
            "  inflating: CNNDataset/test/aa073232b89cd06d.jpg  \n",
            "  inflating: CNNDataset/test/31c0bf8c89fca5e5.jpg  \n",
            "  inflating: CNNDataset/test/0eeb7b2354e16d73.jpg  \n",
            "  inflating: CNNDataset/test/b3afc8866af1e8d1.jpg  \n",
            "  inflating: CNNDataset/test/34b68fd9baa24404.jpg  \n",
            "  inflating: CNNDataset/test/202bc4ed1ad81f8b.jpg  \n",
            "  inflating: CNNDataset/test/7cf55ac1e1553361.jpg  \n",
            "  inflating: CNNDataset/test/c6f31efdddd8bd12.jpg  \n",
            "  inflating: CNNDataset/test/1736e07788834bd2.jpg  \n",
            "  inflating: CNNDataset/test/a7287106de84b405.jpg  \n",
            "  inflating: CNNDataset/test/1409cd2847078968.jpg  \n",
            "  inflating: CNNDataset/test/46e4009df254bd70.jpg  \n",
            "  inflating: CNNDataset/test/19e4437f0c5f5c2a.jpg  \n",
            "  inflating: CNNDataset/test/d0d72976a4ef259b.jpg  \n",
            "  inflating: CNNDataset/test/5eea8a87336506d4.jpg  \n",
            "  inflating: CNNDataset/test/0f2585efba395266.jpg  \n",
            "  inflating: CNNDataset/test/1384b7b72f16c64f.jpg  \n",
            "  inflating: CNNDataset/test/4c81bcf0f5c0e195.jpg  \n",
            "  inflating: CNNDataset/test/1b6c27a199efd739.jpg  \n",
            "  inflating: CNNDataset/test/16f68d119ee4cb6e.jpg  \n",
            "  inflating: CNNDataset/test/b28de23c0556f510.jpg  \n",
            "  inflating: CNNDataset/test/e9cc1045477c195d.jpg  \n",
            "  inflating: CNNDataset/test/f2d74d890a0a90ea.jpg  \n",
            "  inflating: CNNDataset/test/d9c7e178fa11afc8.jpg  \n",
            "  inflating: CNNDataset/test/af3db197b2f78d10.jpg  \n",
            "  inflating: CNNDataset/test/8098506635323131.jpg  \n",
            "  inflating: CNNDataset/test/6b935eeea6290736.jpg  \n",
            "  inflating: CNNDataset/test/366eedd9184ace28.jpg  \n",
            "  inflating: CNNDataset/test/a731395352f2459e.jpg  \n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UauKNtFRqydu"
      },
      "source": [
        "### Import libs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2_iMIh5oIsDa"
      },
      "source": [
        "from __future__ import division\n",
        "from __future__ import print_function\n",
        "from __future__ import absolute_import\n",
        "import random\n",
        "import pprint\n",
        "import sys\n",
        "import time\n",
        "import numpy as np\n",
        "from optparse import OptionParser\n",
        "import pickle\n",
        "import math\n",
        "import cv2\n",
        "import copy\n",
        "from matplotlib import pyplot as plt\n",
        "import tensorflow as tf\n",
        "import pandas as pd\n",
        "import os\n",
        "\n",
        "from sklearn.metrics import average_precision_score\n",
        "\n",
        "from keras import backend as K\n",
        "from tensorflow.keras.optimizers import Adam, SGD, RMSprop\n",
        "from keras.layers import Flatten, Dense, Input, Conv2D, MaxPooling2D, Dropout\n",
        "from keras.layers import GlobalAveragePooling2D, GlobalMaxPooling2D, TimeDistributed\n",
        "from keras.utils.layer_utils import get_source_inputs\n",
        "from keras.utils import layer_utils\n",
        "from keras.utils.data_utils import get_file\n",
        "from keras.metrics import categorical_crossentropy\n",
        "\n",
        "from keras.models import Model\n",
        "from keras.utils import generic_utils\n",
        "from keras.layers import Layer, InputSpec\n",
        "from keras import initializers, regularizers"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WrH5i5mmrDWY"
      },
      "source": [
        "#### Config setting"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DvJm0FFRsyVu"
      },
      "source": [
        "class Config:\n",
        "\n",
        "\tdef __init__(self):\n",
        "\n",
        "\t\t# Print the process or not\n",
        "\t\tself.verbose = True\n",
        "\n",
        "\t\t# Name of base network\n",
        "\t\tself.network = 'vgg'\n",
        "\n",
        "\t\t# Setting for data augmentation\n",
        "\t\tself.use_horizontal_flips = False\n",
        "\t\tself.use_vertical_flips = False\n",
        "\t\tself.rot_90 = False\n",
        "\n",
        "\t\t# Anchor box scales\n",
        "    # Note that if im_size is smaller, anchor_box_scales should be scaled\n",
        "    # Original anchor_box_scales in the paper is [128, 256, 512]\n",
        "\t\tself.anchor_box_scales = [64, 128, 256] \n",
        "\n",
        "\t\t# Anchor box ratios\n",
        "\t\tself.anchor_box_ratios = [[1, 1], [1./math.sqrt(2), 2./math.sqrt(2)], [2./math.sqrt(2), 1./math.sqrt(2)]]\n",
        "\n",
        "\t\t# Size to resize the smallest side of the image\n",
        "\t\t# Original setting in paper is 600. Set to 300 in here to save training time\n",
        "\t\tself.im_size = 300\n",
        "\n",
        "\t\t# image channel-wise mean to subtract\n",
        "\t\tself.img_channel_mean = [103.939, 116.779, 123.68]\n",
        "\t\tself.img_scaling_factor = 1.0\n",
        "\n",
        "\t\t# number of ROIs at once\n",
        "\t\tself.num_rois = 4\n",
        "\n",
        "\t\t# stride at the RPN (this depends on the network configuration)\n",
        "\t\tself.rpn_stride = 16\n",
        "\n",
        "\t\tself.balanced_classes = False\n",
        "\n",
        "\t\t# scaling the stdev\n",
        "\t\tself.std_scaling = 4.0\n",
        "\t\tself.classifier_regr_std = [8.0, 8.0, 4.0, 4.0]\n",
        "\n",
        "\t\t# overlaps for RPN\n",
        "\t\tself.rpn_min_overlap = 0.3\n",
        "\t\tself.rpn_max_overlap = 0.7\n",
        "\n",
        "\t\t# overlaps for classifier ROIs\n",
        "\t\tself.classifier_min_overlap = 0.1\n",
        "\t\tself.classifier_max_overlap = 0.5\n",
        "\n",
        "\t\t# placeholder for the class mapping, automatically generated by the parser\n",
        "\t\tself.class_mapping = None\n",
        "\n",
        "\t\tself.model_path = None"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o0bIjlycyR9_"
      },
      "source": [
        "#### Parser the data from annotation file"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vc89E9uAydTX"
      },
      "source": [
        "def get_data(input_path):\n",
        "\t\"\"\"Parse the data from annotation file\n",
        "\t\n",
        "\tArgs:\n",
        "\t\tinput_path: annotation file path\n",
        "\n",
        "\tReturns:\n",
        "\t\tall_data: list(filepath, width, height, list(bboxes))\n",
        "\t\tclasses_count: dict{key:class_name, value:count_num} \n",
        "\t\t\te.g. {'Car': 2383, 'Mobile phone': 1108, 'Person': 3745}\n",
        "\t\tclass_mapping: dict{key:class_name, value: idx}\n",
        "\t\t\te.g. {'Car': 0, 'Mobile phone': 1, 'Person': 2}\n",
        "\t\"\"\"\n",
        "\tfound_bg = False\n",
        "\tall_imgs = {}\n",
        "\n",
        "\tclasses_count = {}\n",
        "\n",
        "\tclass_mapping = {}\n",
        "\n",
        "\tvisualise = True\n",
        "\n",
        "\ti = 1\n",
        "\t\n",
        "\twith open(input_path,'r') as f:\n",
        "\n",
        "\t\tprint('Parsing annotation files')\n",
        "\n",
        "\t\tfor line in f:\n",
        "\n",
        "\t\t\t# Print process\n",
        "\t\t\tsys.stdout.write('\\r'+'idx=' + str(i))\n",
        "\t\t\ti += 1\n",
        "\n",
        "\t\t\tline_split = line.strip().split(',')\n",
        "\n",
        "\t\t\t# Make sure the info saved in annotation file matching the format (path_filename, x1, y1, x2, y2, class_name)\n",
        "\t\t\t# Note:\n",
        "\t\t\t#\tOne path_filename might has several classes (class_name)\n",
        "\t\t\t#\tx1, y1, x2, y2 are the pixel value of the origial image, not the ratio value\n",
        "\t\t\t#\t(x1, y1) top left coordinates; (x2, y2) bottom right coordinates\n",
        "\t\t\t#   x1,y1-------------------\n",
        "\t\t\t#\t|\t\t\t\t\t\t|\n",
        "\t\t\t#\t|\t\t\t\t\t\t|\n",
        "\t\t\t#\t|\t\t\t\t\t\t|\n",
        "\t\t\t#\t|\t\t\t\t\t\t|\n",
        "\t\t\t#\t---------------------x2,y2\n",
        "\n",
        "\t\t\t(filename,x1,y1,x2,y2,class_name) = line_split\n",
        "\n",
        "\t\t\tif class_name not in classes_count:\n",
        "\t\t\t\tclasses_count[class_name] = 1\n",
        "\t\t\telse:\n",
        "\t\t\t\tclasses_count[class_name] += 1\n",
        "\n",
        "\t\t\tif class_name not in class_mapping:\n",
        "\t\t\t\tif class_name == 'bg' and found_bg == False:\n",
        "\t\t\t\t\tprint('Found class name with special name bg. Will be treated as a background region (this is usually for hard negative mining).')\n",
        "\t\t\t\t\tfound_bg = True\n",
        "\t\t\t\tclass_mapping[class_name] = len(class_mapping)\n",
        "\n",
        "\t\t\tif filename not in all_imgs:\n",
        "\t\t\t\tall_imgs[filename] = {}\n",
        "\t\t\t\t\n",
        "\t\t\t\timg = cv2.imread(filename)\n",
        "\t\t\t\t(rows,cols) = img.shape[:2]\n",
        "\t\t\t\tall_imgs[filename]['filepath'] = filename\n",
        "\t\t\t\tall_imgs[filename]['width'] = cols\n",
        "\t\t\t\tall_imgs[filename]['height'] = rows\n",
        "\t\t\t\tall_imgs[filename]['bboxes'] = []\n",
        "\t\t\t\t# if np.random.randint(0,6) > 0:\n",
        "\t\t\t\t# \tall_imgs[filename]['imageset'] = 'trainval'\n",
        "\t\t\t\t# else:\n",
        "\t\t\t\t# \tall_imgs[filename]['imageset'] = 'test'\n",
        "\n",
        "\t\t\tall_imgs[filename]['bboxes'].append({'class': class_name, 'x1': int(x1), 'x2': int(x2), 'y1': int(y1), 'y2': int(y2)})\n",
        "\n",
        "\n",
        "\t\tall_data = []\n",
        "\t\tfor key in all_imgs:\n",
        "\t\t\tall_data.append(all_imgs[key])\n",
        "\t\t\n",
        "\t\t# make sure the bg class is last in the list\n",
        "\t\tif found_bg:\n",
        "\t\t\tif class_mapping['bg'] != len(class_mapping) - 1:\n",
        "\t\t\t\tkey_to_switch = [key for key in class_mapping.keys() if class_mapping[key] == len(class_mapping)-1][0]\n",
        "\t\t\t\tval_to_switch = class_mapping['bg']\n",
        "\t\t\t\tclass_mapping['bg'] = len(class_mapping) - 1\n",
        "\t\t\t\tclass_mapping[key_to_switch] = val_to_switch\n",
        "\t\t\n",
        "\t\treturn all_data, classes_count, class_mapping"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oFvqGs4acGWl"
      },
      "source": [
        "#### Define ROI Pooling Convolutional Layer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6l32Q85kcMpB"
      },
      "source": [
        "class RoiPoolingConv(Layer):\n",
        "    '''ROI pooling layer for 2D inputs.\n",
        "    See Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,\n",
        "    K. He, X. Zhang, S. Ren, J. Sun\n",
        "    # Arguments\n",
        "        pool_size: int\n",
        "            Size of pooling region to use. pool_size = 7 will result in a 7x7 region.\n",
        "        num_rois: number of regions of interest to be used\n",
        "    # Input shape\n",
        "        list of two 4D tensors [X_img,X_roi] with shape:\n",
        "        X_img:\n",
        "        `(1, rows, cols, channels)`\n",
        "        X_roi:\n",
        "        `(1,num_rois,4)` list of rois, with ordering (x,y,w,h)\n",
        "    # Output shape\n",
        "        3D tensor with shape:\n",
        "        `(1, num_rois, channels, pool_size, pool_size)`\n",
        "    '''\n",
        "    import keras\n",
        "    def __init__(self, pool_size, num_rois, **kwargs):\n",
        "\n",
        "        self.dim_ordering = K.image_data_format()\n",
        "        self.pool_size = pool_size\n",
        "        self.num_rois = num_rois\n",
        "\n",
        "        super(RoiPoolingConv, self).__init__(**kwargs)\n",
        "\n",
        "    def build(self, input_shape):\n",
        "        self.nb_channels = input_shape[0][3]   \n",
        "\n",
        "    def compute_output_shape(self, input_shape):\n",
        "        return None, self.num_rois, self.pool_size, self.pool_size, self.nb_channels\n",
        "\n",
        "    def call(self, x, mask=None):\n",
        "\n",
        "        assert(len(x) == 2)\n",
        "\n",
        "        # x[0] is image with shape (rows, cols, channels)\n",
        "        img = x[0]\n",
        "\n",
        "        # x[1] is roi with shape (num_rois,4) with ordering (x,y,w,h)\n",
        "        rois = x[1]\n",
        "\n",
        "        input_shape = K.shape(img)\n",
        "\n",
        "        outputs = []\n",
        "\n",
        "        for roi_idx in range(self.num_rois):\n",
        "\n",
        "            x = rois[0, roi_idx, 0]\n",
        "            y = rois[0, roi_idx, 1]\n",
        "            w = rois[0, roi_idx, 2]\n",
        "            h = rois[0, roi_idx, 3]\n",
        "\n",
        "            x = K.cast(x, 'int32')\n",
        "            y = K.cast(y, 'int32')\n",
        "            w = K.cast(w, 'int32')\n",
        "            h = K.cast(h, 'int32')\n",
        "\n",
        "            # Resized roi of the image to pooling size (7x7)\n",
        "            rs = tf.image.resize(img[:, y:y+h, x:x+w, :], (self.pool_size, self.pool_size))\n",
        "            outputs.append(rs)\n",
        "                \n",
        "\n",
        "        final_output = K.concatenate(outputs, axis=0)\n",
        "\n",
        "        # Reshape to (1, num_rois, pool_size, pool_size, nb_channels)\n",
        "        # Might be (1, 4, 7, 7, 3)\n",
        "        final_output = K.reshape(final_output, (1, self.num_rois, self.pool_size, self.pool_size, self.nb_channels))\n",
        "\n",
        "        # permute_dimensions is similar to transpose\n",
        "        final_output = K.permute_dimensions(final_output, (0, 1, 2, 3, 4))\n",
        "\n",
        "        return final_output\n",
        "    \n",
        "    \n",
        "    def get_config(self):\n",
        "        config = {'pool_size': self.pool_size,\n",
        "                  'num_rois': self.num_rois}\n",
        "        base_config = super(RoiPoolingConv, self).get_config()\n",
        "        return dict(list(base_config.items()) + list(config.items()))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Mf2taA29RFNs"
      },
      "source": [
        "#### Vgg-16 model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WaBQfl4XRJY3"
      },
      "source": [
        "def get_img_output_length(width, height):\n",
        "    def get_output_length(input_length):\n",
        "        return input_length//16\n",
        "\n",
        "    return get_output_length(width), get_output_length(height)    \n",
        "\n",
        "def nn_base(input_tensor=None, trainable=False):\n",
        "\n",
        "\n",
        "    input_shape = (None, None, 3)\n",
        "\n",
        "    if input_tensor is None:\n",
        "        img_input = Input(shape=input_shape)\n",
        "    else:\n",
        "        if not K.is_keras_tensor(input_tensor):\n",
        "            img_input = Input(tensor=input_tensor, shape=input_shape)\n",
        "        else:\n",
        "            img_input = input_tensor\n",
        "\n",
        "    bn_axis = 3\n",
        "\n",
        "    # Block 1\n",
        "    x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv1')(img_input)\n",
        "    x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x)\n",
        "    x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)\n",
        "\n",
        "    # Block 2\n",
        "    x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x)\n",
        "    x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x)\n",
        "    x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)\n",
        "\n",
        "    # Block 3\n",
        "    x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x)\n",
        "    x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv2')(x)\n",
        "    x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv3')(x)\n",
        "    x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool')(x)\n",
        "\n",
        "    # Block 4\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv1')(x)\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv2')(x)\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv3')(x)\n",
        "    x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool')(x)\n",
        "\n",
        "    # Block 5\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv1')(x)\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv2')(x)\n",
        "    x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv3')(x)\n",
        "    # x = MaxPooling2D((2, 2), strides=(2, 2), name='block5_pool')(x)\n",
        "\n",
        "    return x"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xcOi5MIMVJpU"
      },
      "source": [
        "####  RPN layer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gsuV21vpRczQ"
      },
      "source": [
        "def rpn_layer(base_layers, num_anchors):\n",
        "    \"\"\"Create a rpn layer\n",
        "        Step1: Pass through the feature map from base layer to a 3x3 512 channels convolutional layer\n",
        "                Keep the padding 'same' to preserve the feature map's size\n",
        "        Step2: Pass the step1 to two (1,1) convolutional layer to replace the fully connected layer\n",
        "                classification layer: num_anchors (9 in here) channels for 0, 1 sigmoid activation output\n",
        "                regression layer: num_anchors*4 (36 in here) channels for computing the regression of bboxes with linear activation\n",
        "    Args:\n",
        "        base_layers: vgg in here\n",
        "        num_anchors: 9 in here\n",
        "\n",
        "    Returns:\n",
        "        [x_class, x_regr, base_layers]\n",
        "        x_class: classification for whether it's an object\n",
        "        x_regr: bboxes regression\n",
        "        base_layers: vgg in here\n",
        "    \"\"\"\n",
        "    x = Conv2D(512, (3, 3), padding='same', activation='relu', kernel_initializer='normal', name='rpn_conv1')(base_layers)\n",
        "\n",
        "    x_class = Conv2D(num_anchors, (1, 1), activation='sigmoid', kernel_initializer='uniform', name='rpn_out_class')(x)\n",
        "    x_regr = Conv2D(num_anchors * 4, (1, 1), activation='linear', kernel_initializer='zero', name='rpn_out_regress')(x)\n",
        "\n",
        "    return [x_class, x_regr, base_layers]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0fBt9xNFWsKS"
      },
      "source": [
        "####  Classifier layer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0PKSPLRLWwMz"
      },
      "source": [
        "def classifier_layer(base_layers, input_rois, num_rois, nb_classes = 4):\n",
        "    \"\"\"Create a classifier layer\n",
        "    \n",
        "    Args:\n",
        "        base_layers: vgg\n",
        "        input_rois: `(1,num_rois,4)` list of rois, with ordering (x,y,w,h)\n",
        "        num_rois: number of rois to be processed in one time (4 in here)\n",
        "\n",
        "    Returns:\n",
        "        list(out_class, out_regr)\n",
        "        out_class: classifier layer output\n",
        "        out_regr: regression layer output\n",
        "    \"\"\"\n",
        "\n",
        "    input_shape = (num_rois,7,7,512)\n",
        "\n",
        "    pooling_regions = 7\n",
        "\n",
        "    # out_roi_pool.shape = (1, num_rois, channels, pool_size, pool_size)\n",
        "    # num_rois (4) 7x7 roi pooling\n",
        "    out_roi_pool = RoiPoolingConv(pooling_regions, num_rois)([base_layers, input_rois])\n",
        "\n",
        "    # Flatten the convlutional layer and connected to 2 FC and 2 dropout\n",
        "    out = TimeDistributed(Flatten(name='flatten'))(out_roi_pool)\n",
        "    out = TimeDistributed(Dense(4096, activation='relu', name='fc1'))(out)\n",
        "    out = TimeDistributed(Dropout(0.5))(out)\n",
        "    out = TimeDistributed(Dense(4096, activation='relu', name='fc2'))(out)\n",
        "    out = TimeDistributed(Dropout(0.5))(out)\n",
        "\n",
        "    # There are two output layer\n",
        "    # out_class: softmax acivation function for classify the class name of the object\n",
        "    # out_regr: linear activation function for bboxes coordinates regression\n",
        "    out_class = TimeDistributed(Dense(nb_classes, activation='softmax', kernel_initializer='zero'), name='dense_class_{}'.format(nb_classes))(out)\n",
        "    # note: no regression target for bg class\n",
        "    out_regr = TimeDistributed(Dense(4 * (nb_classes-1), activation='linear', kernel_initializer='zero'), name='dense_regress_{}'.format(nb_classes))(out)\n",
        "\n",
        "    return [out_class, out_regr]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WMev3UMadCzJ"
      },
      "source": [
        "#### Calculate IoU (Intersection of Union)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Jy5iIBYgdCJD"
      },
      "source": [
        "def union(au, bu, area_intersection):\n",
        "\tarea_a = (au[2] - au[0]) * (au[3] - au[1])\n",
        "\tarea_b = (bu[2] - bu[0]) * (bu[3] - bu[1])\n",
        "\tarea_union = area_a + area_b - area_intersection\n",
        "\treturn area_union\n",
        "\n",
        "\n",
        "def intersection(ai, bi):\n",
        "\tx = max(ai[0], bi[0])\n",
        "\ty = max(ai[1], bi[1])\n",
        "\tw = min(ai[2], bi[2]) - x\n",
        "\th = min(ai[3], bi[3]) - y\n",
        "\tif w < 0 or h < 0:\n",
        "\t\treturn 0\n",
        "\treturn w*h\n",
        "\n",
        "\n",
        "def iou(a, b):\n",
        "\t# a and b should be (x1,y1,x2,y2)\n",
        "\n",
        "\tif a[0] >= a[2] or a[1] >= a[3] or b[0] >= b[2] or b[1] >= b[3]:\n",
        "\t\treturn 0.0\n",
        "\n",
        "\tarea_i = intersection(a, b)\n",
        "\tarea_u = union(a, b, area_i)\n",
        "\n",
        "\treturn float(area_i) / float(area_u + 1e-6)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rcRlzqZudKkd"
      },
      "source": [
        "#### Calculate the rpn for all anchors of all images"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "daPsCZtrdK3S"
      },
      "source": [
        "def calc_rpn(C, img_data, width, height, resized_width, resized_height, img_length_calc_function):\n",
        "\t\"\"\"(Important part!) Calculate the rpn for all anchors \n",
        "\t\tIf feature map has shape 38x50=1900, there are 1900x9=17100 potential anchors\n",
        "\t\n",
        "\tArgs:\n",
        "\t\tC: config\n",
        "\t\timg_data: augmented image data\n",
        "\t\twidth: original image width (e.g. 600)\n",
        "\t\theight: original image height (e.g. 800)\n",
        "\t\tresized_width: resized image width according to C.im_size (e.g. 300)\n",
        "\t\tresized_height: resized image height according to C.im_size (e.g. 400)\n",
        "\t\timg_length_calc_function: function to calculate final layer's feature map (of base model) size according to input image size\n",
        "\n",
        "\tReturns:\n",
        "\t\ty_rpn_cls: list(num_bboxes, y_is_box_valid + y_rpn_overlap)\n",
        "\t\t\ty_is_box_valid: 0 or 1 (0 means the box is invalid, 1 means the box is valid)\n",
        "\t\t\ty_rpn_overlap: 0 or 1 (0 means the box is not an object, 1 means the box is an object)\n",
        "\t\ty_rpn_regr: list(num_bboxes, 4*y_rpn_overlap + y_rpn_regr)\n",
        "\t\t\ty_rpn_regr: x1,y1,x2,y2 bunding boxes coordinates\n",
        "\t\"\"\"\n",
        "\tdownscale = float(C.rpn_stride) \n",
        "\tanchor_sizes = C.anchor_box_scales   # 128, 256, 512\n",
        "\tanchor_ratios = C.anchor_box_ratios  # 1:1, 1:2*sqrt(2), 2*sqrt(2):1\n",
        "\tnum_anchors = len(anchor_sizes) * len(anchor_ratios) # 3x3=9\n",
        "\n",
        "\t# calculate the output map size based on the network architecture\n",
        "\t(output_width, output_height) = img_length_calc_function(resized_width, resized_height)\n",
        "\n",
        "\tn_anchratios = len(anchor_ratios)    # 3\n",
        "\t\n",
        "\t# initialise empty output objectives\n",
        "\ty_rpn_overlap = np.zeros((output_height, output_width, num_anchors))\n",
        "\ty_is_box_valid = np.zeros((output_height, output_width, num_anchors))\n",
        "\ty_rpn_regr = np.zeros((output_height, output_width, num_anchors * 4))\n",
        "\n",
        "\tnum_bboxes = len(img_data['bboxes'])\n",
        "\n",
        "\tnum_anchors_for_bbox = np.zeros(num_bboxes).astype(int)\n",
        "\tbest_anchor_for_bbox = -1*np.ones((num_bboxes, 4)).astype(int)\n",
        "\tbest_iou_for_bbox = np.zeros(num_bboxes).astype(np.float32)\n",
        "\tbest_x_for_bbox = np.zeros((num_bboxes, 4)).astype(int)\n",
        "\tbest_dx_for_bbox = np.zeros((num_bboxes, 4)).astype(np.float32)\n",
        "\n",
        "\t# get the GT box coordinates, and resize to account for image resizing\n",
        "\tgta = np.zeros((num_bboxes, 4))\n",
        "\tfor bbox_num, bbox in enumerate(img_data['bboxes']):\n",
        "\t\t# get the GT box coordinates, and resize to account for image resizing\n",
        "\t\tgta[bbox_num, 0] = bbox['x1'] * (resized_width / float(width))\n",
        "\t\tgta[bbox_num, 1] = bbox['x2'] * (resized_width / float(width))\n",
        "\t\tgta[bbox_num, 2] = bbox['y1'] * (resized_height / float(height))\n",
        "\t\tgta[bbox_num, 3] = bbox['y2'] * (resized_height / float(height))\n",
        "\t\n",
        "\t# rpn ground truth\n",
        "\n",
        "\tfor anchor_size_idx in range(len(anchor_sizes)):\n",
        "\t\tfor anchor_ratio_idx in range(n_anchratios):\n",
        "\t\t\tanchor_x = anchor_sizes[anchor_size_idx] * anchor_ratios[anchor_ratio_idx][0]\n",
        "\t\t\tanchor_y = anchor_sizes[anchor_size_idx] * anchor_ratios[anchor_ratio_idx][1]\t\n",
        "\t\t\t\n",
        "\t\t\tfor ix in range(output_width):\t\t\t\t\t\n",
        "\t\t\t\t# x-coordinates of the current anchor box\t\n",
        "\t\t\t\tx1_anc = downscale * (ix + 0.5) - anchor_x / 2\n",
        "\t\t\t\tx2_anc = downscale * (ix + 0.5) + anchor_x / 2\t\n",
        "\t\t\t\t\n",
        "\t\t\t\t# ignore boxes that go across image boundaries\t\t\t\t\t\n",
        "\t\t\t\tif x1_anc < 0 or x2_anc > resized_width:\n",
        "\t\t\t\t\tcontinue\n",
        "\t\t\t\t\t\n",
        "\t\t\t\tfor jy in range(output_height):\n",
        "\n",
        "\t\t\t\t\t# y-coordinates of the current anchor box\n",
        "\t\t\t\t\ty1_anc = downscale * (jy + 0.5) - anchor_y / 2\n",
        "\t\t\t\t\ty2_anc = downscale * (jy + 0.5) + anchor_y / 2\n",
        "\n",
        "\t\t\t\t\t# ignore boxes that go across image boundaries\n",
        "\t\t\t\t\tif y1_anc < 0 or y2_anc > resized_height:\n",
        "\t\t\t\t\t\tcontinue\n",
        "\n",
        "\t\t\t\t\t# bbox_type indicates whether an anchor should be a target\n",
        "\t\t\t\t\t# Initialize with 'negative'\n",
        "\t\t\t\t\tbbox_type = 'neg'\n",
        "\n",
        "\t\t\t\t\t# this is the best IOU for the (x,y) coord and the current anchor\n",
        "\t\t\t\t\t# note that this is different from the best IOU for a GT bbox\n",
        "\t\t\t\t\tbest_iou_for_loc = 0.0\n",
        "\n",
        "\t\t\t\t\tfor bbox_num in range(num_bboxes):\n",
        "\t\t\t\t\t\t\n",
        "\t\t\t\t\t\t# get IOU of the current GT box and the current anchor box\n",
        "\t\t\t\t\t\tcurr_iou = iou([gta[bbox_num, 0], gta[bbox_num, 2], gta[bbox_num, 1], gta[bbox_num, 3]], [x1_anc, y1_anc, x2_anc, y2_anc])\n",
        "\t\t\t\t\t\t# calculate the regression targets if they will be needed\n",
        "\t\t\t\t\t\tif curr_iou > best_iou_for_bbox[bbox_num] or curr_iou > C.rpn_max_overlap:\n",
        "\t\t\t\t\t\t\tcx = (gta[bbox_num, 0] + gta[bbox_num, 1]) / 2.0\n",
        "\t\t\t\t\t\t\tcy = (gta[bbox_num, 2] + gta[bbox_num, 3]) / 2.0\n",
        "\t\t\t\t\t\t\tcxa = (x1_anc + x2_anc)/2.0\n",
        "\t\t\t\t\t\t\tcya = (y1_anc + y2_anc)/2.0\n",
        "\n",
        "\t\t\t\t\t\t\t# x,y are the center point of ground-truth bbox\n",
        "\t\t\t\t\t\t\t# xa,ya are the center point of anchor bbox (xa=downscale * (ix + 0.5); ya=downscale * (iy+0.5))\n",
        "\t\t\t\t\t\t\t# w,h are the width and height of ground-truth bbox\n",
        "\t\t\t\t\t\t\t# wa,ha are the width and height of anchor bboxe\n",
        "\t\t\t\t\t\t\t# tx = (x - xa) / wa\n",
        "\t\t\t\t\t\t\t# ty = (y - ya) / ha\n",
        "\t\t\t\t\t\t\t# tw = log(w / wa)\n",
        "\t\t\t\t\t\t\t# th = log(h / ha)\n",
        "\t\t\t\t\t\t\ttx = (cx - cxa) / (x2_anc - x1_anc)\n",
        "\t\t\t\t\t\t\tty = (cy - cya) / (y2_anc - y1_anc)\n",
        "\t\t\t\t\t\t\ttw = np.log((gta[bbox_num, 1] - gta[bbox_num, 0]) / (x2_anc - x1_anc))\n",
        "\t\t\t\t\t\t\tth = np.log((gta[bbox_num, 3] - gta[bbox_num, 2]) / (y2_anc - y1_anc))\n",
        "\t\t\t\t\t\t\n",
        "\t\t\t\t\t\tif img_data['bboxes'][bbox_num]['class'] != 'bg':\n",
        "\n",
        "\t\t\t\t\t\t\t# all GT boxes should be mapped to an anchor box, so we keep track of which anchor box was best\n",
        "\t\t\t\t\t\t\tif curr_iou > best_iou_for_bbox[bbox_num]:\n",
        "\t\t\t\t\t\t\t\tbest_anchor_for_bbox[bbox_num] = [jy, ix, anchor_ratio_idx, anchor_size_idx]\n",
        "\t\t\t\t\t\t\t\tbest_iou_for_bbox[bbox_num] = curr_iou\n",
        "\t\t\t\t\t\t\t\tbest_x_for_bbox[bbox_num,:] = [x1_anc, x2_anc, y1_anc, y2_anc]\n",
        "\t\t\t\t\t\t\t\tbest_dx_for_bbox[bbox_num,:] = [tx, ty, tw, th]\n",
        "\n",
        "\t\t\t\t\t\t\t# we set the anchor to positive if the IOU is >0.7 (it does not matter if there was another better box, it just indicates overlap)\n",
        "\t\t\t\t\t\t\tif curr_iou > C.rpn_max_overlap:\n",
        "\t\t\t\t\t\t\t\tbbox_type = 'pos'\n",
        "\t\t\t\t\t\t\t\tnum_anchors_for_bbox[bbox_num] += 1\n",
        "\t\t\t\t\t\t\t\t# we update the regression layer target if this IOU is the best for the current (x,y) and anchor position\n",
        "\t\t\t\t\t\t\t\tif curr_iou > best_iou_for_loc:\n",
        "\t\t\t\t\t\t\t\t\tbest_iou_for_loc = curr_iou\n",
        "\t\t\t\t\t\t\t\t\tbest_regr = (tx, ty, tw, th)\n",
        "\n",
        "\t\t\t\t\t\t\t# if the IOU is >0.3 and <0.7, it is ambiguous and no included in the objective\n",
        "\t\t\t\t\t\t\tif C.rpn_min_overlap < curr_iou < C.rpn_max_overlap:\n",
        "\t\t\t\t\t\t\t\t# gray zone between neg and pos\n",
        "\t\t\t\t\t\t\t\tif bbox_type != 'pos':\n",
        "\t\t\t\t\t\t\t\t\tbbox_type = 'neutral'\n",
        "\n",
        "\t\t\t\t\t# turn on or off outputs depending on IOUs\n",
        "\t\t\t\t\tif bbox_type == 'neg':\n",
        "\t\t\t\t\t\ty_is_box_valid[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 1\n",
        "\t\t\t\t\t\ty_rpn_overlap[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 0\n",
        "\t\t\t\t\telif bbox_type == 'neutral':\n",
        "\t\t\t\t\t\ty_is_box_valid[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 0\n",
        "\t\t\t\t\t\ty_rpn_overlap[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 0\n",
        "\t\t\t\t\telif bbox_type == 'pos':\n",
        "\t\t\t\t\t\ty_is_box_valid[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 1\n",
        "\t\t\t\t\t\ty_rpn_overlap[jy, ix, anchor_ratio_idx + n_anchratios * anchor_size_idx] = 1\n",
        "\t\t\t\t\t\tstart = 4 * (anchor_ratio_idx + n_anchratios * anchor_size_idx)\n",
        "\t\t\t\t\t\ty_rpn_regr[jy, ix, start:start+4] = best_regr\n",
        "\n",
        "\t# we ensure that every bbox has at least one positive RPN region\n",
        "\n",
        "\tfor idx in range(num_anchors_for_bbox.shape[0]):\n",
        "\t\tif num_anchors_for_bbox[idx] == 0:\n",
        "\t\t\t# no box with an IOU greater than zero ...\n",
        "\t\t\tif best_anchor_for_bbox[idx, 0] == -1:\n",
        "\t\t\t\tcontinue\n",
        "\t\t\ty_is_box_valid[\n",
        "\t\t\t\tbest_anchor_for_bbox[idx,0], best_anchor_for_bbox[idx,1], best_anchor_for_bbox[idx,2] + n_anchratios *\n",
        "\t\t\t\tbest_anchor_for_bbox[idx,3]] = 1\n",
        "\t\t\ty_rpn_overlap[\n",
        "\t\t\t\tbest_anchor_for_bbox[idx,0], best_anchor_for_bbox[idx,1], best_anchor_for_bbox[idx,2] + n_anchratios *\n",
        "\t\t\t\tbest_anchor_for_bbox[idx,3]] = 1\n",
        "\t\t\tstart = 4 * (best_anchor_for_bbox[idx,2] + n_anchratios * best_anchor_for_bbox[idx,3])\n",
        "\t\t\ty_rpn_regr[\n",
        "\t\t\t\tbest_anchor_for_bbox[idx,0], best_anchor_for_bbox[idx,1], start:start+4] = best_dx_for_bbox[idx, :]\n",
        "\n",
        "\ty_rpn_overlap = np.transpose(y_rpn_overlap, (2, 0, 1))\n",
        "\ty_rpn_overlap = np.expand_dims(y_rpn_overlap, axis=0)\n",
        "\n",
        "\ty_is_box_valid = np.transpose(y_is_box_valid, (2, 0, 1))\n",
        "\ty_is_box_valid = np.expand_dims(y_is_box_valid, axis=0)\n",
        "\n",
        "\ty_rpn_regr = np.transpose(y_rpn_regr, (2, 0, 1))\n",
        "\ty_rpn_regr = np.expand_dims(y_rpn_regr, axis=0)\n",
        "\n",
        "\tpos_locs = np.where(np.logical_and(y_rpn_overlap[0, :, :, :] == 1, y_is_box_valid[0, :, :, :] == 1))\n",
        "\tneg_locs = np.where(np.logical_and(y_rpn_overlap[0, :, :, :] == 0, y_is_box_valid[0, :, :, :] == 1))\n",
        "\n",
        "\tnum_pos = len(pos_locs[0])\n",
        "\n",
        "\t# one issue is that the RPN has many more negative than positive regions, so we turn off some of the negative\n",
        "\t# regions. We also limit it to 256 regions.\n",
        "\tnum_regions = 256\n",
        "\n",
        "\tif len(pos_locs[0]) > num_regions/2:\n",
        "\t\tval_locs = random.sample(range(len(pos_locs[0])), len(pos_locs[0]) - num_regions/2)\n",
        "\t\ty_is_box_valid[0, pos_locs[0][val_locs], pos_locs[1][val_locs], pos_locs[2][val_locs]] = 0\n",
        "\t\tnum_pos = num_regions/2\n",
        "\n",
        "\tif len(neg_locs[0]) + num_pos > num_regions:\n",
        "\t\tval_locs = random.sample(range(len(neg_locs[0])), len(neg_locs[0]) - num_pos)\n",
        "\t\ty_is_box_valid[0, neg_locs[0][val_locs], neg_locs[1][val_locs], neg_locs[2][val_locs]] = 0\n",
        "\n",
        "\ty_rpn_cls = np.concatenate([y_is_box_valid, y_rpn_overlap], axis=1)\n",
        "\ty_rpn_regr = np.concatenate([np.repeat(y_rpn_overlap, 4, axis=1), y_rpn_regr], axis=1)\n",
        "\n",
        "\treturn np.copy(y_rpn_cls), np.copy(y_rpn_regr), num_pos"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3qGAalfJB8zz"
      },
      "source": [
        "#### Get new image size and augment the image"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HKhSFbmB2RTo"
      },
      "source": [
        "def get_new_img_size(width, height, img_min_side=300):\n",
        "\tif width <= height:\n",
        "\t\tf = float(img_min_side) / width\n",
        "\t\tresized_height = int(f * height)\n",
        "\t\tresized_width = img_min_side\n",
        "\telse:\n",
        "\t\tf = float(img_min_side) / height\n",
        "\t\tresized_width = int(f * width)\n",
        "\t\tresized_height = img_min_side\n",
        "\n",
        "\treturn resized_width, resized_height\n",
        "\n",
        "def augment(img_data, config, augment=True):\n",
        "\tassert 'filepath' in img_data\n",
        "\tassert 'bboxes' in img_data\n",
        "\tassert 'width' in img_data\n",
        "\tassert 'height' in img_data\n",
        "\n",
        "\timg_data_aug = copy.deepcopy(img_data)\n",
        "\n",
        "\timg = cv2.imread(img_data_aug['filepath'])\n",
        "\n",
        "\tif augment:\n",
        "\t\trows, cols = img.shape[:2]\n",
        "\n",
        "\t\tif config.use_horizontal_flips and np.random.randint(0, 2) == 0:\n",
        "\t\t\timg = cv2.flip(img, 1)\n",
        "\t\t\tfor bbox in img_data_aug['bboxes']:\n",
        "\t\t\t\tx1 = bbox['x1']\n",
        "\t\t\t\tx2 = bbox['x2']\n",
        "\t\t\t\tbbox['x2'] = cols - x1\n",
        "\t\t\t\tbbox['x1'] = cols - x2\n",
        "\n",
        "\t\tif config.use_vertical_flips and np.random.randint(0, 2) == 0:\n",
        "\t\t\timg = cv2.flip(img, 0)\n",
        "\t\t\tfor bbox in img_data_aug['bboxes']:\n",
        "\t\t\t\ty1 = bbox['y1']\n",
        "\t\t\t\ty2 = bbox['y2']\n",
        "\t\t\t\tbbox['y2'] = rows - y1\n",
        "\t\t\t\tbbox['y1'] = rows - y2\n",
        "\n",
        "\t\tif config.rot_90:\n",
        "\t\t\tangle = np.random.choice([0,90,180,270],1)[0]\n",
        "\t\t\tif angle == 270:\n",
        "\t\t\t\timg = np.transpose(img, (1,0,2))\n",
        "\t\t\t\timg = cv2.flip(img, 0)\n",
        "\t\t\telif angle == 180:\n",
        "\t\t\t\timg = cv2.flip(img, -1)\n",
        "\t\t\telif angle == 90:\n",
        "\t\t\t\timg = np.transpose(img, (1,0,2))\n",
        "\t\t\t\timg = cv2.flip(img, 1)\n",
        "\t\t\telif angle == 0:\n",
        "\t\t\t\tpass\n",
        "\n",
        "\t\t\tfor bbox in img_data_aug['bboxes']:\n",
        "\t\t\t\tx1 = bbox['x1']\n",
        "\t\t\t\tx2 = bbox['x2']\n",
        "\t\t\t\ty1 = bbox['y1']\n",
        "\t\t\t\ty2 = bbox['y2']\n",
        "\t\t\t\tif angle == 270:\n",
        "\t\t\t\t\tbbox['x1'] = y1\n",
        "\t\t\t\t\tbbox['x2'] = y2\n",
        "\t\t\t\t\tbbox['y1'] = cols - x2\n",
        "\t\t\t\t\tbbox['y2'] = cols - x1\n",
        "\t\t\t\telif angle == 180:\n",
        "\t\t\t\t\tbbox['x2'] = cols - x1\n",
        "\t\t\t\t\tbbox['x1'] = cols - x2\n",
        "\t\t\t\t\tbbox['y2'] = rows - y1\n",
        "\t\t\t\t\tbbox['y1'] = rows - y2\n",
        "\t\t\t\telif angle == 90:\n",
        "\t\t\t\t\tbbox['x1'] = rows - y2\n",
        "\t\t\t\t\tbbox['x2'] = rows - y1\n",
        "\t\t\t\t\tbbox['y1'] = x1\n",
        "\t\t\t\t\tbbox['y2'] = x2        \n",
        "\t\t\t\telif angle == 0:\n",
        "\t\t\t\t\tpass\n",
        "\n",
        "\timg_data_aug['width'] = img.shape[1]\n",
        "\timg_data_aug['height'] = img.shape[0]\n",
        "\treturn img_data_aug, img"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0712o8CXkyh1"
      },
      "source": [
        "#### Generate the ground_truth anchors"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TvsEv3RIk0cF"
      },
      "source": [
        "def get_anchor_gt(all_img_data, C, img_length_calc_function, mode='train'):\n",
        "\t\"\"\" Yield the ground-truth anchors as Y (labels)\n",
        "\t\t\n",
        "\tArgs:\n",
        "\t\tall_img_data: list(filepath, width, height, list(bboxes))\n",
        "\t\tC: config\n",
        "\t\timg_length_calc_function: function to calculate final layer's feature map (of base model) size according to input image size\n",
        "\t\tmode: 'train' or 'test'; 'train' mode need augmentation\n",
        "\n",
        "\tReturns:\n",
        "\t\tx_img: image data after resized and scaling (smallest size = 300px)\n",
        "\t\tY: [y_rpn_cls, y_rpn_regr]\n",
        "\t\timg_data_aug: augmented image data (original image with augmentation)\n",
        "\t\tdebug_img: show image for debug\n",
        "\t\tnum_pos: show number of positive anchors for debug\n",
        "\t\"\"\"\n",
        "\twhile True:\n",
        "\n",
        "\t\tfor img_data in all_img_data:\n",
        "\t\t\ttry:\n",
        "\n",
        "\t\t\t\t# read in image, and optionally add augmentation\n",
        "\n",
        "\t\t\t\tif mode == 'train':\n",
        "\t\t\t\t\timg_data_aug, x_img = augment(img_data, C, augment=True)\n",
        "\t\t\t\telse:\n",
        "\t\t\t\t\timg_data_aug, x_img = augment(img_data, C, augment=False)\n",
        "\n",
        "\t\t\t\t(width, height) = (img_data_aug['width'], img_data_aug['height'])\n",
        "\t\t\t\t(rows, cols, _) = x_img.shape\n",
        "\n",
        "\t\t\t\tassert cols == width\n",
        "\t\t\t\tassert rows == height\n",
        "\n",
        "\t\t\t\t# get image dimensions for resizing\n",
        "\t\t\t\t(resized_width, resized_height) = get_new_img_size(width, height, C.im_size)\n",
        "\n",
        "\t\t\t\t# resize the image so that smalles side is length = 300px\n",
        "\t\t\t\tx_img = cv2.resize(x_img, (resized_width, resized_height), interpolation=cv2.INTER_CUBIC)\n",
        "\t\t\t\tdebug_img = x_img.copy()\n",
        "\n",
        "\t\t\t\ttry:\n",
        "\t\t\t\t\ty_rpn_cls, y_rpn_regr, num_pos = calc_rpn(C, img_data_aug, width, height, resized_width, resized_height, img_length_calc_function)\n",
        "\t\t\t\texcept:\n",
        "\t\t\t\t\tcontinue\n",
        "\n",
        "\t\t\t\t# Zero-center by mean pixel, and preprocess image\n",
        "\n",
        "\t\t\t\tx_img = x_img[:,:, (2, 1, 0)]  # BGR -> RGB\n",
        "\t\t\t\tx_img = x_img.astype(np.float32)\n",
        "\t\t\t\tx_img[:, :, 0] -= C.img_channel_mean[0]\n",
        "\t\t\t\tx_img[:, :, 1] -= C.img_channel_mean[1]\n",
        "\t\t\t\tx_img[:, :, 2] -= C.img_channel_mean[2]\n",
        "\t\t\t\tx_img /= C.img_scaling_factor\n",
        "\n",
        "\t\t\t\tx_img = np.transpose(x_img, (2, 0, 1))\n",
        "\t\t\t\tx_img = np.expand_dims(x_img, axis=0)\n",
        "\n",
        "\t\t\t\ty_rpn_regr[:, y_rpn_regr.shape[1]//2:, :, :] *= C.std_scaling\n",
        "\n",
        "\t\t\t\tx_img = np.transpose(x_img, (0, 2, 3, 1))\n",
        "\t\t\t\ty_rpn_cls = np.transpose(y_rpn_cls, (0, 2, 3, 1))\n",
        "\t\t\t\ty_rpn_regr = np.transpose(y_rpn_regr, (0, 2, 3, 1))\n",
        "\n",
        "\t\t\t\tyield np.copy(x_img), [np.copy(y_rpn_cls), np.copy(y_rpn_regr)], img_data_aug, debug_img, num_pos\n",
        "\n",
        "\t\t\texcept Exception as e:\n",
        "\t\t\t\tprint(e)\n",
        "\t\t\t\tcontinue"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FZAAMEH4uqu9"
      },
      "source": [
        "#### Define loss functions for all four outputs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CyLxnL4_uvmr"
      },
      "source": [
        "lambda_rpn_regr = 1.0\n",
        "lambda_rpn_class = 1.0\n",
        "\n",
        "lambda_cls_regr = 1.0\n",
        "lambda_cls_class = 1.0\n",
        "\n",
        "epsilon = 1e-4"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tvGfH6m3yu0_"
      },
      "source": [
        "def rpn_loss_regr(num_anchors):\n",
        "    \"\"\"Loss function for rpn regression\n",
        "    Args:\n",
        "        num_anchors: number of anchors (9 in here)\n",
        "    Returns:\n",
        "        Smooth L1 loss function \n",
        "                           0.5*x*x (if x_abs < 1)\n",
        "                           x_abx - 0.5 (otherwise)\n",
        "    \"\"\"\n",
        "    def rpn_loss_regr_fixed_num(y_true, y_pred):\n",
        "\n",
        "        # x is the difference between true value and predicted vaue\n",
        "        x = y_true[:, :, :, 4 * num_anchors:] - y_pred\n",
        "\n",
        "        # absolute value of x\n",
        "        x_abs = K.abs(x)\n",
        "\n",
        "        # If x_abs <= 1.0, x_bool = 1\n",
        "        x_bool = K.cast(K.less_equal(x_abs, 1.0), tf.float32)\n",
        "\n",
        "        return lambda_rpn_regr * K.sum(\n",
        "            y_true[:, :, :, :4 * num_anchors] * (x_bool * (0.5 * x * x) + (1 - x_bool) * (x_abs - 0.5))) / K.sum(epsilon + y_true[:, :, :, :4 * num_anchors])\n",
        "\n",
        "    return rpn_loss_regr_fixed_num\n",
        "\n",
        "\n",
        "def rpn_loss_cls(num_anchors):\n",
        "    \"\"\"Loss function for rpn classification\n",
        "    Args:\n",
        "        num_anchors: number of anchors (9 in here)\n",
        "        y_true[:, :, :, :9]: [0,1,0,0,0,0,0,1,0] means only the second and the eighth box is valid which contains pos or neg anchor => isValid\n",
        "        y_true[:, :, :, 9:]: [0,1,0,0,0,0,0,0,0] means the second box is pos and eighth box is negative\n",
        "    Returns:\n",
        "        lambda * sum((binary_crossentropy(isValid*y_pred,y_true))) / N\n",
        "    \"\"\"\n",
        "    def rpn_loss_cls_fixed_num(y_true, y_pred):\n",
        "\n",
        "            return lambda_rpn_class * K.sum(y_true[:, :, :, :num_anchors] * K.binary_crossentropy(y_pred[:, :, :, :], y_true[:, :, :, num_anchors:])) / K.sum(epsilon + y_true[:, :, :, :num_anchors])\n",
        "\n",
        "    return rpn_loss_cls_fixed_num\n",
        "\n",
        "\n",
        "def class_loss_regr(num_classes):\n",
        "    \"\"\"Loss function for rpn regression\n",
        "    Args:\n",
        "        num_anchors: number of anchors (9 in here)\n",
        "    Returns:\n",
        "        Smooth L1 loss function \n",
        "                           0.5*x*x (if x_abs < 1)\n",
        "                           x_abx - 0.5 (otherwise)\n",
        "    \"\"\"\n",
        "    def class_loss_regr_fixed_num(y_true, y_pred):\n",
        "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
        "        x_abs = K.abs(x)\n",
        "        x_bool = K.cast(K.less_equal(x_abs, 1.0), 'float32')\n",
        "        return lambda_cls_regr * K.sum(y_true[:, :, :4*num_classes] * (x_bool * (0.5 * x * x) + (1 - x_bool) * (x_abs - 0.5))) / K.sum(epsilon + y_true[:, :, :4*num_classes])\n",
        "    return class_loss_regr_fixed_num\n",
        "\n",
        "\n",
        "def class_loss_cls(y_true, y_pred):\n",
        "    return lambda_cls_class * K.mean(categorical_crossentropy(y_true[0, :, :], y_pred[0, :, :]))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5cX0N4VDl4zS"
      },
      "source": [
        "def non_max_suppression_fast(boxes, probs, overlap_thresh=0.9, max_boxes=300):\n",
        "    # code used from here: http://www.pyimagesearch.com/2015/02/16/faster-non-maximum-suppression-python/\n",
        "    # if there are no boxes, return an empty list\n",
        "\n",
        "    # Process explanation:\n",
        "    #   Step 1: Sort the probs list\n",
        "    #   Step 2: Find the larget prob 'Last' in the list and save it to the pick list\n",
        "    #   Step 3: Calculate the IoU with 'Last' box and other boxes in the list. If the IoU is larger than overlap_threshold, delete the box from list\n",
        "    #   Step 4: Repeat step 2 and step 3 until there is no item in the probs list \n",
        "    if len(boxes) == 0:\n",
        "        return []\n",
        "\n",
        "    # grab the coordinates of the bounding boxes\n",
        "    x1 = boxes[:, 0]\n",
        "    y1 = boxes[:, 1]\n",
        "    x2 = boxes[:, 2]\n",
        "    y2 = boxes[:, 3]\n",
        "\n",
        "    np.testing.assert_array_less(x1, x2)\n",
        "    np.testing.assert_array_less(y1, y2)\n",
        "\n",
        "    # if the bounding boxes integers, convert them to floats --\n",
        "    # this is important since we'll be doing a bunch of divisions\n",
        "    if boxes.dtype.kind == \"i\":\n",
        "        boxes = boxes.astype(\"float\")\n",
        "\n",
        "    # initialize the list of picked indexes\t\n",
        "    pick = []\n",
        "\n",
        "    # calculate the areas\n",
        "    area = (x2 - x1) * (y2 - y1)\n",
        "\n",
        "    # sort the bounding boxes \n",
        "    idxs = np.argsort(probs)\n",
        "\n",
        "    # keep looping while some indexes still remain in the indexes\n",
        "    # list\n",
        "    while len(idxs) > 0:\n",
        "        # grab the last index in the indexes list and add the\n",
        "        # index value to the list of picked indexes\n",
        "        last = len(idxs) - 1\n",
        "        i = idxs[last]\n",
        "        pick.append(i)\n",
        "\n",
        "        # find the intersection\n",
        "\n",
        "        xx1_int = np.maximum(x1[i], x1[idxs[:last]])\n",
        "        yy1_int = np.maximum(y1[i], y1[idxs[:last]])\n",
        "        xx2_int = np.minimum(x2[i], x2[idxs[:last]])\n",
        "        yy2_int = np.minimum(y2[i], y2[idxs[:last]])\n",
        "\n",
        "        ww_int = np.maximum(0, xx2_int - xx1_int)\n",
        "        hh_int = np.maximum(0, yy2_int - yy1_int)\n",
        "\n",
        "        area_int = ww_int * hh_int\n",
        "\n",
        "        # find the union\n",
        "        area_union = area[i] + area[idxs[:last]] - area_int\n",
        "\n",
        "        # compute the ratio of overlap\n",
        "        overlap = area_int/(area_union + 1e-6)\n",
        "\n",
        "        # delete all indexes from the index list that have\n",
        "        idxs = np.delete(idxs, np.concatenate(([last],\n",
        "            np.where(overlap > overlap_thresh)[0])))\n",
        "\n",
        "        if len(pick) >= max_boxes:\n",
        "            break\n",
        "\n",
        "    # return only the bounding boxes that were picked using the integer data type\n",
        "    boxes = boxes[pick].astype(\"int\")\n",
        "    probs = probs[pick]\n",
        "    return boxes, probs\n",
        "\n",
        "def apply_regr_np(X, T):\n",
        "    \"\"\"Apply regression layer to all anchors in one feature map\n",
        "\n",
        "    Args:\n",
        "        X: shape=(4, 18, 25) the current anchor type for all points in the feature map\n",
        "        T: regression layer shape=(4, 18, 25)\n",
        "\n",
        "    Returns:\n",
        "        X: regressed position and size for current anchor\n",
        "    \"\"\"\n",
        "    try:\n",
        "        x = X[0, :, :]\n",
        "        y = X[1, :, :]\n",
        "        w = X[2, :, :]\n",
        "        h = X[3, :, :]\n",
        "\n",
        "        tx = T[0, :, :]\n",
        "        ty = T[1, :, :]\n",
        "        tw = T[2, :, :]\n",
        "        th = T[3, :, :]\n",
        "\n",
        "        cx = x + w/2.\n",
        "        cy = y + h/2.\n",
        "        cx1 = tx * w + cx\n",
        "        cy1 = ty * h + cy\n",
        "\n",
        "        w1 = np.exp(tw.astype(np.float64)) * w\n",
        "        h1 = np.exp(th.astype(np.float64)) * h\n",
        "        x1 = cx1 - w1/2.\n",
        "        y1 = cy1 - h1/2.\n",
        "\n",
        "        x1 = np.round(x1)\n",
        "        y1 = np.round(y1)\n",
        "        w1 = np.round(w1)\n",
        "        h1 = np.round(h1)\n",
        "        return np.stack([x1, y1, w1, h1])\n",
        "    except Exception as e:\n",
        "        print(e)\n",
        "        return X\n",
        "    \n",
        "def apply_regr(x, y, w, h, tx, ty, tw, th):\n",
        "    # Apply regression to x, y, w and h\n",
        "    try:\n",
        "        cx = x + w/2.\n",
        "        cy = y + h/2.\n",
        "        cx1 = tx * w + cx\n",
        "        cy1 = ty * h + cy\n",
        "        w1 = math.exp(tw) * w\n",
        "        h1 = math.exp(th) * h\n",
        "        x1 = cx1 - w1/2.\n",
        "        y1 = cy1 - h1/2.\n",
        "        x1 = int(round(x1))\n",
        "        y1 = int(round(y1))\n",
        "        w1 = int(round(w1))\n",
        "        h1 = int(round(h1))\n",
        "\n",
        "        return x1, y1, w1, h1\n",
        "\n",
        "    except ValueError:\n",
        "        return x, y, w, h\n",
        "    except OverflowError:\n",
        "        return x, y, w, h\n",
        "    except Exception as e:\n",
        "        print(e)\n",
        "        return x, y, w, h\n",
        "\n",
        "def calc_iou(R, img_data, C, class_mapping):\n",
        "    \"\"\"Converts from (x1,y1,x2,y2) to (x,y,w,h) format\n",
        "\n",
        "    Args:\n",
        "        R: bboxes, probs\n",
        "    \"\"\"\n",
        "    bboxes = img_data['bboxes']\n",
        "    (width, height) = (img_data['width'], img_data['height'])\n",
        "    # get image dimensions for resizing\n",
        "    (resized_width, resized_height) = get_new_img_size(width, height, C.im_size)\n",
        "\n",
        "    gta = np.zeros((len(bboxes), 4))\n",
        "\n",
        "    for bbox_num, bbox in enumerate(bboxes):\n",
        "        # get the GT box coordinates, and resize to account for image resizing\n",
        "        # gta[bbox_num, 0] = (40 * (600 / 800)) / 16 = int(round(1.875)) = 2 (x in feature map)\n",
        "        gta[bbox_num, 0] = int(round(bbox['x1'] * (resized_width / float(width))/C.rpn_stride))\n",
        "        gta[bbox_num, 1] = int(round(bbox['x2'] * (resized_width / float(width))/C.rpn_stride))\n",
        "        gta[bbox_num, 2] = int(round(bbox['y1'] * (resized_height / float(height))/C.rpn_stride))\n",
        "        gta[bbox_num, 3] = int(round(bbox['y2'] * (resized_height / float(height))/C.rpn_stride))\n",
        "\n",
        "    x_roi = []\n",
        "    y_class_num = []\n",
        "    y_class_regr_coords = []\n",
        "    y_class_regr_label = []\n",
        "    IoUs = [] # for debugging only\n",
        "\n",
        "    # R.shape[0]: number of bboxes (=300 from non_max_suppression)\n",
        "    for ix in range(R.shape[0]):\n",
        "        (x1, y1, x2, y2) = R[ix, :]\n",
        "        x1 = int(round(x1))\n",
        "        y1 = int(round(y1))\n",
        "        x2 = int(round(x2))\n",
        "        y2 = int(round(y2))\n",
        "\n",
        "        best_iou = 0.0\n",
        "        best_bbox = -1\n",
        "        # Iterate through all the ground-truth bboxes to calculate the iou\n",
        "        for bbox_num in range(len(bboxes)):\n",
        "            curr_iou = iou([gta[bbox_num, 0], gta[bbox_num, 2], gta[bbox_num, 1], gta[bbox_num, 3]], [x1, y1, x2, y2])\n",
        "\n",
        "            # Find out the corresponding ground-truth bbox_num with larget iou\n",
        "            if curr_iou > best_iou:\n",
        "                best_iou = curr_iou\n",
        "                best_bbox = bbox_num\n",
        "\n",
        "        if best_iou < C.classifier_min_overlap:\n",
        "                continue\n",
        "        else:\n",
        "            w = x2 - x1\n",
        "            h = y2 - y1\n",
        "            x_roi.append([x1, y1, w, h])\n",
        "            IoUs.append(best_iou)\n",
        "\n",
        "            if C.classifier_min_overlap <= best_iou < C.classifier_max_overlap:\n",
        "                # hard negative example\n",
        "                cls_name = 'bg'\n",
        "            elif C.classifier_max_overlap <= best_iou:\n",
        "                cls_name = bboxes[best_bbox]['class']\n",
        "                cxg = (gta[best_bbox, 0] + gta[best_bbox, 1]) / 2.0\n",
        "                cyg = (gta[best_bbox, 2] + gta[best_bbox, 3]) / 2.0\n",
        "\n",
        "                cx = x1 + w / 2.0\n",
        "                cy = y1 + h / 2.0\n",
        "\n",
        "                tx = (cxg - cx) / float(w)\n",
        "                ty = (cyg - cy) / float(h)\n",
        "                tw = np.log((gta[best_bbox, 1] - gta[best_bbox, 0]) / float(w))\n",
        "                th = np.log((gta[best_bbox, 3] - gta[best_bbox, 2]) / float(h))\n",
        "            else:\n",
        "                print('roi = {}'.format(best_iou))\n",
        "                raise RuntimeError\n",
        "\n",
        "        class_num = class_mapping[cls_name]\n",
        "        class_label = len(class_mapping) * [0]\n",
        "        class_label[class_num] = 1\n",
        "        y_class_num.append(copy.deepcopy(class_label))\n",
        "        coords = [0] * 4 * (len(class_mapping) - 1)\n",
        "        labels = [0] * 4 * (len(class_mapping) - 1)\n",
        "        if cls_name != 'bg':\n",
        "            label_pos = 4 * class_num\n",
        "            sx, sy, sw, sh = C.classifier_regr_std\n",
        "            coords[label_pos:4+label_pos] = [sx*tx, sy*ty, sw*tw, sh*th]\n",
        "            labels[label_pos:4+label_pos] = [1, 1, 1, 1]\n",
        "            y_class_regr_coords.append(copy.deepcopy(coords))\n",
        "            y_class_regr_label.append(copy.deepcopy(labels))\n",
        "        else:\n",
        "            y_class_regr_coords.append(copy.deepcopy(coords))\n",
        "            y_class_regr_label.append(copy.deepcopy(labels))\n",
        "\n",
        "    if len(x_roi) == 0:\n",
        "        return None, None, None, None\n",
        "\n",
        "    # bboxes that iou > C.classifier_min_overlap for all gt bboxes in 300 non_max_suppression bboxes\n",
        "    X = np.array(x_roi)\n",
        "    # one hot code for bboxes from above => x_roi (X)\n",
        "    Y1 = np.array(y_class_num)\n",
        "    # corresponding labels and corresponding gt bboxes\n",
        "    Y2 = np.concatenate([np.array(y_class_regr_label),np.array(y_class_regr_coords)],axis=1)\n",
        "\n",
        "    return np.expand_dims(X, axis=0), np.expand_dims(Y1, axis=0), np.expand_dims(Y2, axis=0), IoUs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vT6X-fqJ1RSl"
      },
      "source": [
        "def rpn_to_roi(rpn_layer, regr_layer, C, dim_ordering, use_regr=True, max_boxes=300,overlap_thresh=0.9):\n",
        "\t\"\"\"Convert rpn layer to roi bboxes\n",
        "\n",
        "\tArgs: (num_anchors = 9)\n",
        "\t\trpn_layer: output layer for rpn classification \n",
        "\t\t\tshape (1, feature_map.height, feature_map.width, num_anchors)\n",
        "\t\t\tMight be (1, 18, 25, 18) if resized image is 400 width and 300\n",
        "\t\tregr_layer: output layer for rpn regression\n",
        "\t\t\tshape (1, feature_map.height, feature_map.width, num_anchors)\n",
        "\t\t\tMight be (1, 18, 25, 72) if resized image is 400 width and 300\n",
        "\t\tC: config\n",
        "\t\tuse_regr: Wether to use bboxes regression in rpn\n",
        "\t\tmax_boxes: max bboxes number for non-max-suppression (NMS)\n",
        "\t\toverlap_thresh: If iou in NMS is larger than this threshold, drop the box\n",
        "\n",
        "\tReturns:\n",
        "\t\tresult: boxes from non-max-suppression (shape=(300, 4))\n",
        "\t\t\tboxes: coordinates for bboxes (on the feature map)\n",
        "\t\"\"\"\n",
        "\tregr_layer = regr_layer / C.std_scaling\n",
        "\n",
        "\tanchor_sizes = C.anchor_box_scales   # (3 in here)\n",
        "\tanchor_ratios = C.anchor_box_ratios  # (3 in here)\n",
        "\n",
        "\tassert rpn_layer.shape[0] == 1\n",
        "\n",
        "\t(rows, cols) = rpn_layer.shape[1:3]\n",
        "\n",
        "\tcurr_layer = 0\n",
        "\n",
        "\t# A.shape = (4, feature_map.height, feature_map.width, num_anchors) \n",
        "\t# Might be (4, 18, 25, 18) if resized image is 400 width and 300\n",
        "\t# A is the coordinates for 9 anchors for every point in the feature map \n",
        "\t# => all 18x25x9=4050 anchors cooridnates\n",
        "\tA = np.zeros((4, rpn_layer.shape[1], rpn_layer.shape[2], rpn_layer.shape[3]))\n",
        "\n",
        "\tfor anchor_size in anchor_sizes:\n",
        "\t\tfor anchor_ratio in anchor_ratios:\n",
        "\t\t\t# anchor_x = (128 * 1) / 16 = 8  => width of current anchor\n",
        "\t\t\t# anchor_y = (128 * 2) / 16 = 16 => height of current anchor\n",
        "\t\t\tanchor_x = (anchor_size * anchor_ratio[0])/C.rpn_stride\n",
        "\t\t\tanchor_y = (anchor_size * anchor_ratio[1])/C.rpn_stride\n",
        "\t\t\t\n",
        "\t\t\t# curr_layer: 0~8 (9 anchors)\n",
        "\t\t\t# the Kth anchor of all position in the feature map (9th in total)\n",
        "\t\t\tregr = regr_layer[0, :, :, 4 * curr_layer:4 * curr_layer + 4] # shape => (18, 25, 4)\n",
        "\t\t\tregr = np.transpose(regr, (2, 0, 1)) # shape => (4, 18, 25)\n",
        "\n",
        "\t\t\t# Create 18x25 mesh grid\n",
        "\t\t\t# For every point in x, there are all the y points and vice versa\n",
        "\t\t\t# X.shape = (18, 25)\n",
        "\t\t\t# Y.shape = (18, 25)\n",
        "\t\t\tX, Y = np.meshgrid(np.arange(cols),np. arange(rows))\n",
        "\n",
        "\t\t\t# Calculate anchor position and size for each feature map point\n",
        "\t\t\tA[0, :, :, curr_layer] = X - anchor_x/2 # Top left x coordinate\n",
        "\t\t\tA[1, :, :, curr_layer] = Y - anchor_y/2 # Top left y coordinate\n",
        "\t\t\tA[2, :, :, curr_layer] = anchor_x       # width of current anchor\n",
        "\t\t\tA[3, :, :, curr_layer] = anchor_y       # height of current anchor\n",
        "\n",
        "\t\t\t# Apply regression to x, y, w and h if there is rpn regression layer\n",
        "\t\t\tif use_regr:\n",
        "\t\t\t\tA[:, :, :, curr_layer] = apply_regr_np(A[:, :, :, curr_layer], regr)\n",
        "\n",
        "\t\t\t# Avoid width and height exceeding 1\n",
        "\t\t\tA[2, :, :, curr_layer] = np.maximum(1, A[2, :, :, curr_layer])\n",
        "\t\t\tA[3, :, :, curr_layer] = np.maximum(1, A[3, :, :, curr_layer])\n",
        "\n",
        "\t\t\t# Convert (x, y , w, h) to (x1, y1, x2, y2)\n",
        "\t\t\t# x1, y1 is top left coordinate\n",
        "\t\t\t# x2, y2 is bottom right coordinate\n",
        "\t\t\tA[2, :, :, curr_layer] += A[0, :, :, curr_layer]\n",
        "\t\t\tA[3, :, :, curr_layer] += A[1, :, :, curr_layer]\n",
        "\n",
        "\t\t\t# Avoid bboxes drawn outside the feature map\n",
        "\t\t\tA[0, :, :, curr_layer] = np.maximum(0, A[0, :, :, curr_layer])\n",
        "\t\t\tA[1, :, :, curr_layer] = np.maximum(0, A[1, :, :, curr_layer])\n",
        "\t\t\tA[2, :, :, curr_layer] = np.minimum(cols-1, A[2, :, :, curr_layer])\n",
        "\t\t\tA[3, :, :, curr_layer] = np.minimum(rows-1, A[3, :, :, curr_layer])\n",
        "\n",
        "\t\t\tcurr_layer += 1\n",
        "\n",
        "\tall_boxes = np.reshape(A.transpose((0, 3, 1, 2)), (4, -1)).transpose((1, 0))  # shape=(4050, 4)\n",
        "\tall_probs = rpn_layer.transpose((0, 3, 1, 2)).reshape((-1))                   # shape=(4050,)\n",
        "\n",
        "\tx1 = all_boxes[:, 0]\n",
        "\ty1 = all_boxes[:, 1]\n",
        "\tx2 = all_boxes[:, 2]\n",
        "\ty2 = all_boxes[:, 3]\n",
        "\n",
        "\t# Find out the bboxes which is illegal and delete them from bboxes list\n",
        "\tidxs = np.where((x1 - x2 >= 0) | (y1 - y2 >= 0))\n",
        "\n",
        "\tall_boxes = np.delete(all_boxes, idxs, 0)\n",
        "\tall_probs = np.delete(all_probs, idxs, 0)\n",
        "\n",
        "\t# Apply non_max_suppression\n",
        "\t# Only extract the bboxes. Don't need rpn probs in the later process\n",
        "\tresult = non_max_suppression_fast(all_boxes, all_probs, overlap_thresh=overlap_thresh, max_boxes=max_boxes)[0]\n",
        "\n",
        "\treturn result"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Kk14GTaNmqoo"
      },
      "source": [
        "\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oNsi6HtyJPSb"
      },
      "source": [
        "\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TVmMqXE5x70U"
      },
      "source": [
        "### Start training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C66bqGuOq7w6"
      },
      "source": [
        "base_path = '/content/CNNDataset'\n",
        "\n",
        "train_path =  '/content/CNNDataset/annotation.txt' # Training data (annotation file)\n",
        "\n",
        "num_rois = 4 # Number of RoIs to process at once.\n",
        "\n",
        "# Augmentation flag\n",
        "horizontal_flips = True # Augment with horizontal flips in training. \n",
        "vertical_flips = True   # Augment with vertical flips in training. \n",
        "rot_90 = True           # Augment with 90 degree rotations in training. \n",
        "\n",
        "output_weight_path = os.path.join(base_path, 'model/model_frcnn_vgg.hdf5')\n",
        "\n",
        "record_path = os.path.join(base_path, 'model/record.csv') # Record data (used to save the losses, classification accuracy and mean average precision)\n",
        "\n",
        "base_weight_path = os.path.join(base_path, 'model/vgg16_weights_tf_dim_ordering_tf_kernels.h5')\n",
        "\n",
        "config_output_filename = os.path.join(base_path, 'model_vgg_config.pickle')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J3oAmbbEutH0"
      },
      "source": [
        "# Create the config\n",
        "C = Config()\n",
        "\n",
        "C.use_horizontal_flips = horizontal_flips\n",
        "C.use_vertical_flips = vertical_flips\n",
        "C.rot_90 = rot_90\n",
        "\n",
        "C.record_path = record_path\n",
        "C.model_path = output_weight_path\n",
        "C.num_rois = num_rois\n",
        "\n",
        "C.base_net_weights = base_weight_path"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yiEaAmb-x-so",
        "outputId": "41ab3976-cd75-4a1f-e69d-101f32f4d7f2"
      },
      "source": [
        "#--------------------------------------------------------#\n",
        "# This step will spend some time to load the data        #\n",
        "#--------------------------------------------------------#\n",
        "st = time.time()\n",
        "train_imgs, classes_count, class_mapping = get_data(train_path)\n",
        "print()\n",
        "print('Spend %0.2f mins to load the data' % ((time.time()-st)/60) )"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Parsing annotation files\n",
            "idx=831\n",
            "Spend 0.39 mins to load the data\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "x-nuSdC56GsK",
        "outputId": "d10e1b7e-045f-4a3a-9469-f7cf87ff4ede"
      },
      "source": [
        "if 'bg' not in classes_count:\n",
        "\tclasses_count['bg'] = 0\n",
        "\tclass_mapping['bg'] = len(class_mapping)\n",
        "# e.g.\n",
        "#    classes_count: {'Car': 2383, 'Mobile phone': 1108, 'Person': 3745, 'bg': 0}\n",
        "#    class_mapping: {'Person': 0, 'Car': 1, 'Mobile phone': 2, 'bg': 3}\n",
        "C.class_mapping = class_mapping\n",
        "\n",
        "print('Training images per class:')\n",
        "pprint.pprint(classes_count)\n",
        "print('Num classes (including bg) = {}'.format(len(classes_count)))\n",
        "print(class_mapping)\n",
        "\n",
        "# Save the configuration\n",
        "with open(config_output_filename, 'wb') as config_f:\n",
        "\tpickle.dump(C,config_f)\n",
        "\tprint('Config has been written to {}, and can be loaded when testing to ensure correct results'.format(config_output_filename))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Training images per class:\n",
            "{'Car': 575, 'Truck': 256, 'bg': 0}\n",
            "Num classes (including bg) = 3\n",
            "{'Car': 0, 'Truck': 1, 'bg': 2}\n",
            "Config has been written to /content/CNNDataset/model_vgg_config.pickle, and can be loaded when testing to ensure correct results\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LFlq36Sx4F4O",
        "outputId": "a5e37349-5d26-4f21-90c8-1ad15de94f31"
      },
      "source": [
        "# Shuffle the images with seed\n",
        "random.seed(1)\n",
        "random.shuffle(train_imgs)\n",
        "\n",
        "print('Num train samples (images) {}'.format(len(train_imgs)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Num train samples (images) 320\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GXIV1uXyBo3v"
      },
      "source": [
        "# Get train data generator which generate X, Y, image_data\n",
        "data_gen_train = get_anchor_gt(train_imgs, C, get_img_output_length, mode='train')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y_yM5jkKqM1G"
      },
      "source": [
        "#### Explore 'data_gen_train'\n",
        "\n",
        "data_gen_train is an **generator**, so we get the data by calling **next(data_gen_train)**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nIDnio1UlRHi"
      },
      "source": [
        "X, Y, image_data, debug_img, debug_num_pos = next(data_gen_train)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 848
        },
        "id": "dZXoJ2e3l2Ey",
        "outputId": "5ce4fcc5-f323-457c-9b85-03ca4d6027a4"
      },
      "source": [
        "print('Original image: height=%d width=%d'%(image_data['height'], image_data['width']))\n",
        "print('Resized image:  height=%d width=%d C.im_size=%d'%(X.shape[1], X.shape[2], C.im_size))\n",
        "print('Feature map size: height=%d width=%d C.rpn_stride=%d'%(Y[0].shape[1], Y[0].shape[2], C.rpn_stride))\n",
        "print(X.shape)\n",
        "print(str(len(Y))+\" includes 'y_rpn_cls' and 'y_rpn_regr'\")\n",
        "print('Shape of y_rpn_cls {}'.format(Y[0].shape))\n",
        "print('Shape of y_rpn_regr {}'.format(Y[1].shape))\n",
        "print(image_data)\n",
        "\n",
        "print('Number of positive anchors for this image: %d' % (debug_num_pos))\n",
        "if debug_num_pos==0:\n",
        "    gt_x1, gt_x2 = image_data['bboxes'][0]['x1']*(X.shape[2]/image_data['height']), image_data['bboxes'][0]['x2']*(X.shape[2]/image_data['height'])\n",
        "    gt_y1, gt_y2 = image_data['bboxes'][0]['y1']*(X.shape[1]/image_data['width']), image_data['bboxes'][0]['y2']*(X.shape[1]/image_data['width'])\n",
        "    gt_x1, gt_y1, gt_x2, gt_y2 = int(gt_x1), int(gt_y1), int(gt_x2), int(gt_y2)\n",
        "\n",
        "    img = debug_img.copy()\n",
        "    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n",
        "    color = (0, 255, 0)\n",
        "    cv2.putText(img, 'gt bbox', (gt_x1, gt_y1-5), cv2.FONT_HERSHEY_DUPLEX, 0.7, color, 1)\n",
        "    cv2.rectangle(img, (gt_x1, gt_y1), (gt_x2, gt_y2), color, 2)\n",
        "    cv2.circle(img, (int((gt_x1+gt_x2)/2), int((gt_y1+gt_y2)/2)), 3, color, -1)\n",
        "\n",
        "    plt.grid()\n",
        "    plt.imshow(img)\n",
        "    plt.show()\n",
        "else:\n",
        "    cls = Y[0][0]\n",
        "    pos_cls = np.where(cls==1)\n",
        "    print(pos_cls)\n",
        "    regr = Y[1][0]\n",
        "    pos_regr = np.where(regr==1)\n",
        "    print(pos_regr)\n",
        "    print('y_rpn_cls for possible pos anchor: {}'.format(cls[pos_cls[0][0],pos_cls[1][0],:]))\n",
        "    print('y_rpn_regr for positive anchor: {}'.format(regr[pos_regr[0][0],pos_regr[1][0],:]))\n",
        "\n",
        "    gt_x1, gt_x2 = image_data['bboxes'][0]['x1']*(X.shape[2]/image_data['width']), image_data['bboxes'][0]['x2']*(X.shape[2]/image_data['width'])\n",
        "    gt_y1, gt_y2 = image_data['bboxes'][0]['y1']*(X.shape[1]/image_data['height']), image_data['bboxes'][0]['y2']*(X.shape[1]/image_data['height'])\n",
        "    gt_x1, gt_y1, gt_x2, gt_y2 = int(gt_x1), int(gt_y1), int(gt_x2), int(gt_y2)\n",
        "\n",
        "    img = debug_img.copy()\n",
        "    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n",
        "    color = (0, 255, 0)\n",
        "    #   cv2.putText(img, 'gt bbox', (gt_x1, gt_y1-5), cv2.FONT_HERSHEY_DUPLEX, 0.7, color, 1)\n",
        "    cv2.rectangle(img, (gt_x1, gt_y1), (gt_x2, gt_y2), color, 2)\n",
        "    cv2.circle(img, (int((gt_x1+gt_x2)/2), int((gt_y1+gt_y2)/2)), 3, color, -1)\n",
        "\n",
        "    # Add text\n",
        "    textLabel = 'gt bbox'\n",
        "    (retval,baseLine) = cv2.getTextSize(textLabel,cv2.FONT_HERSHEY_COMPLEX,0.5,1)\n",
        "    textOrg = (gt_x1, gt_y1+5)\n",
        "    cv2.rectangle(img, (textOrg[0] - 5, textOrg[1]+baseLine - 5), (textOrg[0]+retval[0] + 5, textOrg[1]-retval[1] - 5), (0, 0, 0), 2)\n",
        "    cv2.rectangle(img, (textOrg[0] - 5,textOrg[1]+baseLine - 5), (textOrg[0]+retval[0] + 5, textOrg[1]-retval[1] - 5), (255, 255, 255), -1)\n",
        "    cv2.putText(img, textLabel, textOrg, cv2.FONT_HERSHEY_DUPLEX, 0.5, (0, 0, 0), 1)\n",
        "\n",
        "    # Draw positive anchors according to the y_rpn_regr\n",
        "    for i in range(debug_num_pos):\n",
        "\n",
        "        color = (100+i*(155/4), 0, 100+i*(155/4))\n",
        "\n",
        "        idx = pos_regr[2][i*4]/4\n",
        "        anchor_size = C.anchor_box_scales[int(idx/3)]\n",
        "        anchor_ratio = C.anchor_box_ratios[2-int((idx+1)%3)]\n",
        "\n",
        "        center = (pos_regr[1][i*4]*C.rpn_stride, pos_regr[0][i*4]*C.rpn_stride)\n",
        "        print('Center position of positive anchor: ', center)\n",
        "        cv2.circle(img, center, 3, color, -1)\n",
        "        anc_w, anc_h = anchor_size*anchor_ratio[0], anchor_size*anchor_ratio[1]\n",
        "        cv2.rectangle(img, (center[0]-int(anc_w/2), center[1]-int(anc_h/2)), (center[0]+int(anc_w/2), center[1]+int(anc_h/2)), color, 2)\n",
        "#         cv2.putText(img, 'pos anchor bbox '+str(i+1), (center[0]-int(anc_w/2), center[1]-int(anc_h/2)-5), cv2.FONT_HERSHEY_DUPLEX, 0.5, color, 1)\n",
        "\n",
        "print('Green bboxes is ground-truth bbox. Others are positive anchors')\n",
        "plt.figure(figsize=(8,8))\n",
        "plt.grid()\n",
        "plt.imshow(img)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Original image: height=1536 width=2048\n",
            "Resized image:  height=300 width=400 C.im_size=300\n",
            "Feature map size: height=18 width=25 C.rpn_stride=16\n",
            "(1, 300, 400, 3)\n",
            "2 includes 'y_rpn_cls' and 'y_rpn_regr'\n",
            "Shape of y_rpn_cls (1, 18, 25, 18)\n",
            "Shape of y_rpn_regr (1, 18, 25, 72)\n",
            "{'filepath': '/content/CNNDataset/train/c4d05cc4ebb07dc4.jpg', 'width': 2048, 'height': 1536, 'bboxes': [{'class': 'Truck', 'x1': 640, 'x2': 2046, 'y1': 248, 'y2': 954}]}\n",
            "Number of positive anchors for this image: 1\n",
            "(array([ 8,  8, 13]), array([13, 13, 14]), array([ 8, 17,  0]))\n",
            "(array([8, 8, 8, 8]), array([13, 13, 13, 13]), array([32, 33, 34, 35]))\n",
            "y_rpn_cls for possible pos anchor: [0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 1.]\n",
            "y_rpn_regr for positive anchor: [ 0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          1.          1.          1.          1.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.          0.          0.          0.\n",
            "  0.          0.          0.51159936 -0.4113856  -1.1056056  -1.08857226]\n",
            "Center position of positive anchor:  (208, 128)\n",
            "Green bboxes is ground-truth bbox. Others are positive anchors\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i4XSyIoubCMY"
      },
      "source": [
        "#### Build the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jODipXFDnDJ0"
      },
      "source": [
        "input_shape_img = (None, None, 3)\n",
        "\n",
        "img_input = Input(shape=input_shape_img)\n",
        "roi_input = Input(shape=(None, 4))\n",
        "\n",
        "# define the base network (VGG here, can be Resnet50, Inception, etc)\n",
        "shared_layers = nn_base(img_input, trainable=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "udTeQMVhfSzw",
        "outputId": "0137dca5-f82e-4161-ed9f-3bca334db2c0"
      },
      "source": [
        "# define the RPN, built on the base layers\n",
        "num_anchors = len(C.anchor_box_scales) * len(C.anchor_box_ratios) # 9\n",
        "rpn = rpn_layer(shared_layers, num_anchors)\n",
        "\n",
        "classifier = classifier_layer(shared_layers, roi_input, C.num_rois, nb_classes=len(classes_count))\n",
        "\n",
        "model_rpn = Model(img_input, rpn[:2])\n",
        "model_classifier = Model([img_input, roi_input], classifier)\n",
        "\n",
        "# this is a model that holds both the RPN and the classifier, used to load/save weights for the models\n",
        "model_all = Model([img_input, roi_input], rpn[:2] + classifier)\n",
        "\n",
        "# Because the google colab can only run the session several hours one time (then you need to connect again), \n",
        "# we need to save the model and load the model to continue training\n",
        "if not os.path.isfile(C.model_path):\n",
        "    #If this is the begin of the training, load the pre-traind base network such as vgg-16\n",
        "    try:\n",
        "        print('This is the first time of your training')\n",
        "        print('loading weights from {}'.format(C.base_net_weights))\n",
        "        model_rpn.load_weights(C.base_net_weights, by_name=True)\n",
        "        model_classifier.load_weights(C.base_net_weights, by_name=True)\n",
        "    except:\n",
        "        print('Could not load pretrained model weights. Weights can be found in the keras application folder \\\n",
        "            https://github.com/fchollet/keras/tree/master/keras/applications')\n",
        "    \n",
        "    # Create the record.csv file to record losses, acc and mAP\n",
        "    record_df = pd.DataFrame(columns=['mean_overlapping_bboxes', 'class_acc', 'loss_rpn_cls', 'loss_rpn_regr', 'loss_class_cls', 'loss_class_regr', 'curr_loss', 'elapsed_time', 'mAP'])\n",
        "else:\n",
        "    # If this is a continued training, load the trained model from before\n",
        "    print('Continue training based on previous trained model')\n",
        "    print('Loading weights from {}'.format(C.model_path))\n",
        "    model_rpn.load_weights(C.model_path, by_name=True)\n",
        "    model_classifier.load_weights(C.model_path, by_name=True)\n",
        "    \n",
        "    # Load the records\n",
        "    record_df = pd.read_csv(record_path)\n",
        "\n",
        "    r_mean_overlapping_bboxes = record_df['mean_overlapping_bboxes']\n",
        "    r_class_acc = record_df['class_acc']\n",
        "    r_loss_rpn_cls = record_df['loss_rpn_cls']\n",
        "    r_loss_rpn_regr = record_df['loss_rpn_regr']\n",
        "    r_loss_class_cls = record_df['loss_class_cls']\n",
        "    r_loss_class_regr = record_df['loss_class_regr']\n",
        "    r_curr_loss = record_df['curr_loss']\n",
        "    r_elapsed_time = record_df['elapsed_time']\n",
        "    r_mAP = record_df['mAP']\n",
        "\n",
        "    print('Already train %dK batches'% (len(record_df)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "This is the first time of your training\n",
            "loading weights from /content/CNNDataset/model/vgg16_weights_tf_dim_ordering_tf_kernels.h5\n",
            "Could not load pretrained model weights. Weights can be found in the keras application folder             https://github.com/fchollet/keras/tree/master/keras/applications\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-ULrg0V1soIR",
        "outputId": "159baf94-9fd9-4b70-df8b-5661b143ead0"
      },
      "source": [
        "optimizer = Adam(lr=1e-5)\n",
        "optimizer_classifier = Adam(lr=1e-5)\n",
        "model_rpn.compile(optimizer=optimizer, loss=[rpn_loss_cls(num_anchors), rpn_loss_regr(num_anchors)])\n",
        "model_classifier.compile(optimizer=optimizer_classifier, loss=[class_loss_cls, class_loss_regr(len(classes_count)-1)], metrics={'dense_class_{}'.format(len(classes_count)): 'accuracy'})\n",
        "model_all.compile(optimizer='sgd', loss='mae')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n",
            "  super(Adam, self).__init__(name, **kwargs)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Qz2BYzL6sqfu"
      },
      "source": [
        "# Training setting\n",
        "total_epochs = len(record_df)\n",
        "r_epochs = len(record_df)\n",
        "\n",
        "epoch_length = 1000\n",
        "num_epochs = 7\n",
        "iter_num = 0\n",
        "\n",
        "total_epochs += num_epochs\n",
        "\n",
        "losses = np.zeros((epoch_length, 5))\n",
        "rpn_accuracy_rpn_monitor = []\n",
        "rpn_accuracy_for_epoch = []\n",
        "\n",
        "if len(record_df)==0:\n",
        "    best_loss = np.Inf\n",
        "else:\n",
        "    best_loss = np.min(r_curr_loss)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JDysEDQA2DUz",
        "outputId": "c5c23299-b981-44ab-a2e4-fa90dee2bd71"
      },
      "source": [
        "print(len(record_df))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dRXtd5W30DRN",
        "outputId": "4d766c34-b0bc-4406-910e-2753ab5581ad"
      },
      "source": [
        "start_time = time.time()\n",
        "for epoch_num in range(num_epochs):\n",
        "\n",
        "    progbar = generic_utils.Progbar(epoch_length)\n",
        "    print('Epoch {}/{}'.format(r_epochs + 1, total_epochs))\n",
        "    \n",
        "    r_epochs += 1\n",
        "\n",
        "    while True:\n",
        "        try:\n",
        "\n",
        "            if len(rpn_accuracy_rpn_monitor) == epoch_length and C.verbose:\n",
        "                mean_overlapping_bboxes = float(sum(rpn_accuracy_rpn_monitor))/len(rpn_accuracy_rpn_monitor)\n",
        "                rpn_accuracy_rpn_monitor = []\n",
        "#                 print('Average number of overlapping bounding boxes from RPN = {} for {} previous iterations'.format(mean_overlapping_bboxes, epoch_length))\n",
        "                if mean_overlapping_bboxes == 0:\n",
        "                    print('RPN is not producing bounding boxes that overlap the ground truth boxes. Check RPN settings or keep training.')\n",
        "\n",
        "            # Generate X (x_img) and label Y ([y_rpn_cls, y_rpn_regr])\n",
        "            X, Y, img_data, debug_img, debug_num_pos = next(data_gen_train)\n",
        "\n",
        "            # Train rpn model and get loss value [_, loss_rpn_cls, loss_rpn_regr]\n",
        "            loss_rpn = model_rpn.train_on_batch(X, Y)\n",
        "\n",
        "            # Get predicted rpn from rpn model [rpn_cls, rpn_regr]\n",
        "            P_rpn = model_rpn.predict_on_batch(X)\n",
        "\n",
        "            # R: bboxes (shape=(300,4))\n",
        "            # Convert rpn layer to roi bboxes\n",
        "            R = rpn_to_roi(P_rpn[0], P_rpn[1], C, K.image_data_format(), use_regr=True, overlap_thresh=0.7, max_boxes=300)\n",
        "            \n",
        "            # note: calc_iou converts from (x1,y1,x2,y2) to (x,y,w,h) format\n",
        "            # X2: bboxes that iou > C.classifier_min_overlap for all gt bboxes in 300 non_max_suppression bboxes\n",
        "            # Y1: one hot code for bboxes from above => x_roi (X)\n",
        "            # Y2: corresponding labels and corresponding gt bboxes\n",
        "            X2, Y1, Y2, IouS = calc_iou(R, img_data, C, class_mapping)\n",
        "\n",
        "            # If X2 is None means there are no matching bboxes\n",
        "            if X2 is None:\n",
        "                rpn_accuracy_rpn_monitor.append(0)\n",
        "                rpn_accuracy_for_epoch.append(0)\n",
        "                continue\n",
        "            \n",
        "            # Find out the positive anchors and negative anchors\n",
        "            neg_samples = np.where(Y1[0, :, -1] == 1)\n",
        "            pos_samples = np.where(Y1[0, :, -1] == 0)\n",
        "\n",
        "            if len(neg_samples) > 0:\n",
        "                neg_samples = neg_samples[0]\n",
        "            else:\n",
        "                neg_samples = []\n",
        "\n",
        "            if len(pos_samples) > 0:\n",
        "                pos_samples = pos_samples[0]\n",
        "            else:\n",
        "                pos_samples = []\n",
        "\n",
        "            rpn_accuracy_rpn_monitor.append(len(pos_samples))\n",
        "            rpn_accuracy_for_epoch.append((len(pos_samples)))\n",
        "\n",
        "            if C.num_rois > 1:\n",
        "                # If number of positive anchors is larger than 4//2 = 2, randomly choose 2 pos samples\n",
        "                if len(pos_samples) < C.num_rois//2:\n",
        "                    selected_pos_samples = pos_samples.tolist()\n",
        "                else:\n",
        "                    selected_pos_samples = np.random.choice(pos_samples, C.num_rois//2, replace=False).tolist()\n",
        "                \n",
        "                # Randomly choose (num_rois - num_pos) neg samples\n",
        "                try:\n",
        "                    selected_neg_samples = np.random.choice(neg_samples, C.num_rois - len(selected_pos_samples), replace=False).tolist()\n",
        "                except:\n",
        "                    selected_neg_samples = np.random.choice(neg_samples, C.num_rois - len(selected_pos_samples), replace=True).tolist()\n",
        "                \n",
        "                # Save all the pos and neg samples in sel_samples\n",
        "                sel_samples = selected_pos_samples + selected_neg_samples\n",
        "            else:\n",
        "                # in the extreme case where num_rois = 1, we pick a random pos or neg sample\n",
        "                selected_pos_samples = pos_samples.tolist()\n",
        "                selected_neg_samples = neg_samples.tolist()\n",
        "                if np.random.randint(0, 2):\n",
        "                    sel_samples = random.choice(neg_samples)\n",
        "                else:\n",
        "                    sel_samples = random.choice(pos_samples)\n",
        "\n",
        "            # training_data: [X, X2[:, sel_samples, :]]\n",
        "            # labels: [Y1[:, sel_samples, :], Y2[:, sel_samples, :]]\n",
        "            #  X                     => img_data resized image\n",
        "            #  X2[:, sel_samples, :] => num_rois (4 in here) bboxes which contains selected neg and pos\n",
        "            #  Y1[:, sel_samples, :] => one hot encode for num_rois bboxes which contains selected neg and pos\n",
        "            #  Y2[:, sel_samples, :] => labels and gt bboxes for num_rois bboxes which contains selected neg and pos\n",
        "            loss_class = model_classifier.train_on_batch([X, X2[:, sel_samples, :]], [Y1[:, sel_samples, :], Y2[:, sel_samples, :]])\n",
        "\n",
        "            losses[iter_num, 0] = loss_rpn[1]\n",
        "            losses[iter_num, 1] = loss_rpn[2]\n",
        "\n",
        "            losses[iter_num, 2] = loss_class[1]\n",
        "            losses[iter_num, 3] = loss_class[2]\n",
        "            losses[iter_num, 4] = loss_class[3]\n",
        "\n",
        "            iter_num += 1\n",
        "\n",
        "            progbar.update(iter_num, [('rpn_cls', np.mean(losses[:iter_num, 0])), ('rpn_regr', np.mean(losses[:iter_num, 1])),\n",
        "                                      ('final_cls', np.mean(losses[:iter_num, 2])), ('final_regr', np.mean(losses[:iter_num, 3]))])\n",
        "\n",
        "            if iter_num == epoch_length:\n",
        "                loss_rpn_cls = np.mean(losses[:, 0])\n",
        "                loss_rpn_regr = np.mean(losses[:, 1])\n",
        "                loss_class_cls = np.mean(losses[:, 2])\n",
        "                loss_class_regr = np.mean(losses[:, 3])\n",
        "                class_acc = np.mean(losses[:, 4])\n",
        "\n",
        "                mean_overlapping_bboxes = float(sum(rpn_accuracy_for_epoch)) / len(rpn_accuracy_for_epoch)\n",
        "                rpn_accuracy_for_epoch = []\n",
        "\n",
        "                if C.verbose:\n",
        "                    print('Mean number of bounding boxes from RPN overlapping ground truth boxes: {}'.format(mean_overlapping_bboxes))\n",
        "                    print('Classifier accuracy for bounding boxes from RPN: {}'.format(class_acc))\n",
        "                    print('Loss RPN classifier: {}'.format(loss_rpn_cls))\n",
        "                    print('Loss RPN regression: {}'.format(loss_rpn_regr))\n",
        "                    print('Loss Detector classifier: {}'.format(loss_class_cls))\n",
        "                    print('Loss Detector regression: {}'.format(loss_class_regr))\n",
        "                    print('Total loss: {}'.format(loss_rpn_cls + loss_rpn_regr + loss_class_cls + loss_class_regr))\n",
        "                    print('Elapsed time: {}'.format(time.time() - start_time))\n",
        "                    elapsed_time = (time.time()-start_time)/60\n",
        "\n",
        "                curr_loss = loss_rpn_cls + loss_rpn_regr + loss_class_cls + loss_class_regr\n",
        "                iter_num = 0\n",
        "                start_time = time.time()\n",
        "\n",
        "                if curr_loss < best_loss:\n",
        "                    if C.verbose:\n",
        "                        print('Total loss decreased from {} to {}, saving weights'.format(best_loss,curr_loss))\n",
        "                    best_loss = curr_loss\n",
        "                    model_all.save_weights(C.model_path)\n",
        "\n",
        "                new_row = {'mean_overlapping_bboxes':round(mean_overlapping_bboxes, 3), \n",
        "                           'class_acc':round(class_acc, 3), \n",
        "                           'loss_rpn_cls':round(loss_rpn_cls, 3), \n",
        "                           'loss_rpn_regr':round(loss_rpn_regr, 3), \n",
        "                           'loss_class_cls':round(loss_class_cls, 3), \n",
        "                           'loss_class_regr':round(loss_class_regr, 3), \n",
        "                           'curr_loss':round(curr_loss, 3), \n",
        "                           'elapsed_time':round(elapsed_time, 3), \n",
        "                           'mAP': 0}\n",
        "\n",
        "                record_df = record_df.append(new_row, ignore_index=True)\n",
        "                record_df.to_csv(record_path, index=0)\n",
        "\n",
        "                break\n",
        "\n",
        "        except Exception as e:\n",
        "            print('Exception: {}'.format(e))\n",
        "            continue\n",
        "\n",
        "print('Training complete, exiting.')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[1;30;43mStreaming output truncated to the last 5000 lines.\u001b[0m\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 628/1000 [=================>............] - ETA: 5:26 - rpn_cls: 2.4885 - rpn_regr: 0.3277 - final_cls: 0.9695 - final_regr: 0.4188Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 648/1000 [==================>...........] - ETA: 5:07 - rpn_cls: 2.4892 - rpn_regr: 0.3279 - final_cls: 0.9691 - final_regr: 0.4188Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 674/1000 [===================>..........] - ETA: 4:44 - rpn_cls: 2.4902 - rpn_regr: 0.3281 - final_cls: 0.9685 - final_regr: 0.4186Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 680/1000 [===================>..........] - ETA: 4:39 - rpn_cls: 2.4905 - rpn_regr: 0.3281 - final_cls: 0.9684 - final_regr: 0.4186Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 685/1000 [===================>..........] - ETA: 4:35 - rpn_cls: 2.4907 - rpn_regr: 0.3281 - final_cls: 0.9683 - final_regr: 0.4185Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 704/1000 [====================>.........] - ETA: 4:18 - rpn_cls: 2.4922 - rpn_regr: 0.3283 - final_cls: 0.9680 - final_regr: 0.4184Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 714/1000 [====================>.........] - ETA: 4:09 - rpn_cls: 2.4930 - rpn_regr: 0.3284 - final_cls: 0.9678 - final_regr: 0.4183Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 720/1000 [====================>.........] - ETA: 4:04 - rpn_cls: 2.4934 - rpn_regr: 0.3284 - final_cls: 0.9677 - final_regr: 0.4182Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 749/1000 [=====================>........] - ETA: 3:38 - rpn_cls: 2.4948 - rpn_regr: 0.3287 - final_cls: 0.9673 - final_regr: 0.4180Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 752/1000 [=====================>........] - ETA: 3:35 - rpn_cls: 2.4950 - rpn_regr: 0.3287 - final_cls: 0.9672 - final_regr: 0.4180Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 767/1000 [======================>.......] - ETA: 3:22 - rpn_cls: 2.4955 - rpn_regr: 0.3288 - final_cls: 0.9670 - final_regr: 0.4179Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 784/1000 [======================>.......] - ETA: 3:07 - rpn_cls: 2.4961 - rpn_regr: 0.3289 - final_cls: 0.9668 - final_regr: 0.4178Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 799/1000 [======================>.......] - ETA: 2:54 - rpn_cls: 2.4965 - rpn_regr: 0.3290 - final_cls: 0.9666 - final_regr: 0.4177Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 800/1000 [=======================>......] - ETA: 2:54 - rpn_cls: 2.4966 - rpn_regr: 0.3290 - final_cls: 0.9666 - final_regr: 0.4177Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 812/1000 [=======================>......] - ETA: 2:43 - rpn_cls: 2.4971 - rpn_regr: 0.3291 - final_cls: 0.9665 - final_regr: 0.4177Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 825/1000 [=======================>......] - ETA: 2:32 - rpn_cls: 2.4976 - rpn_regr: 0.3292 - final_cls: 0.9664 - final_regr: 0.4176Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 847/1000 [========================>.....] - ETA: 2:13 - rpn_cls: 2.4985 - rpn_regr: 0.3295 - final_cls: 0.9662 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 850/1000 [========================>.....] - ETA: 2:10 - rpn_cls: 2.4986 - rpn_regr: 0.3296 - final_cls: 0.9662 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 852/1000 [========================>.....] - ETA: 2:09 - rpn_cls: 2.4987 - rpn_regr: 0.3296 - final_cls: 0.9661 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 854/1000 [========================>.....] - ETA: 2:07 - rpn_cls: 2.4988 - rpn_regr: 0.3296 - final_cls: 0.9661 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 864/1000 [========================>.....] - ETA: 1:58 - rpn_cls: 2.4992 - rpn_regr: 0.3298 - final_cls: 0.9661 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 866/1000 [========================>.....] - ETA: 1:57 - rpn_cls: 2.4993 - rpn_regr: 0.3298 - final_cls: 0.9661 - final_regr: 0.4175Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 885/1000 [=========================>....] - ETA: 1:40 - rpn_cls: 2.5000 - rpn_regr: 0.3300 - final_cls: 0.9660 - final_regr: 0.4174Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 890/1000 [=========================>....] - ETA: 1:36 - rpn_cls: 2.5003 - rpn_regr: 0.3301 - final_cls: 0.9659 - final_regr: 0.4174Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 895/1000 [=========================>....] - ETA: 1:31 - rpn_cls: 2.5005 - rpn_regr: 0.3301 - final_cls: 0.9659 - final_regr: 0.4173Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 907/1000 [==========================>...] - ETA: 1:21 - rpn_cls: 2.5010 - rpn_regr: 0.3303 - final_cls: 0.9658 - final_regr: 0.4173Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 917/1000 [==========================>...] - ETA: 1:12 - rpn_cls: 2.5013 - rpn_regr: 0.3303 - final_cls: 0.9657 - final_regr: 0.4172Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 937/1000 [===========================>..] - ETA: 54s - rpn_cls: 2.5017 - rpn_regr: 0.3305 - final_cls: 0.9656 - final_regr: 0.4171Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 963/1000 [===========================>..] - ETA: 32s - rpn_cls: 2.5022 - rpn_regr: 0.3306 - final_cls: 0.9653 - final_regr: 0.4170Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 970/1000 [============================>.] - ETA: 26s - rpn_cls: 2.5024 - rpn_regr: 0.3307 - final_cls: 0.9653 - final_regr: 0.4169Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 976/1000 [============================>.] - ETA: 20s - rpn_cls: 2.5025 - rpn_regr: 0.3307 - final_cls: 0.9652 - final_regr: 0.4169Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 978/1000 [============================>.] - ETA: 19s - rpn_cls: 2.5026 - rpn_regr: 0.3307 - final_cls: 0.9652 - final_regr: 0.4169Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 986/1000 [============================>.] - ETA: 12s - rpn_cls: 2.5026 - rpn_regr: 0.3307 - final_cls: 0.9651 - final_regr: 0.4168Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 993/1000 [============================>.] - ETA: 6s - rpn_cls: 2.5027 - rpn_regr: 0.3308 - final_cls: 0.9650 - final_regr: 0.4168Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 876s 875ms/step - rpn_cls: 2.5028 - rpn_regr: 0.3308 - final_cls: 0.9650 - final_regr: 0.4167\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.042152466367712\n",
            "Classifier accuracy for bounding boxes from RPN: 0.54375\n",
            "Loss RPN classifier: 2.520514255403359\n",
            "Loss RPN regression: 0.3356346467640251\n",
            "Loss Detector classifier: 0.9560902360081672\n",
            "Loss Detector regression: 0.40741085129557175\n",
            "Total loss: 4.219649989471123\n",
            "Elapsed time: 878.5104503631592\n",
            "Total loss decreased from 4.426501114243958 to 4.219649989471123, saving weights\n",
            "Epoch 10/40\n",
            "  33/1000 [..............................] - ETA: 13:08 - rpn_cls: 2.0279 - rpn_regr: 0.3032 - final_cls: 0.9748 - final_regr: 0.4018Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  42/1000 [>.............................] - ETA: 13:06 - rpn_cls: 2.0032 - rpn_regr: 0.3167 - final_cls: 0.9695 - final_regr: 0.4030Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  70/1000 [=>............................] - ETA: 13:12 - rpn_cls: 2.0065 - rpn_regr: 0.3255 - final_cls: 0.9725 - final_regr: 0.4122Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  74/1000 [=>............................] - ETA: 13:20 - rpn_cls: 2.0052 - rpn_regr: 0.3256 - final_cls: 0.9726 - final_regr: 0.4133Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  89/1000 [=>............................] - ETA: 13:16 - rpn_cls: 2.0039 - rpn_regr: 0.3274 - final_cls: 0.9723 - final_regr: 0.4166Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  90/1000 [=>............................] - ETA: 13:28 - rpn_cls: 2.0037 - rpn_regr: 0.3277 - final_cls: 0.9722 - final_regr: 0.4167Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 108/1000 [==>...........................] - ETA: 13:12 - rpn_cls: 2.0381 - rpn_regr: 0.3319 - final_cls: 0.9724 - final_regr: 0.4196Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 115/1000 [==>...........................] - ETA: 13:13 - rpn_cls: 2.0554 - rpn_regr: 0.3337 - final_cls: 0.9721 - final_regr: 0.4201Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 118/1000 [==>...........................] - ETA: 13:13 - rpn_cls: 2.0613 - rpn_regr: 0.3343 - final_cls: 0.9719 - final_regr: 0.4201Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 136/1000 [===>..........................] - ETA: 13:00 - rpn_cls: 2.0929 - rpn_regr: 0.3384 - final_cls: 0.9701 - final_regr: 0.4199Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 145/1000 [===>..........................] - ETA: 12:56 - rpn_cls: 2.1084 - rpn_regr: 0.3403 - final_cls: 0.9693 - final_regr: 0.4194Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 155/1000 [===>..........................] - ETA: 12:46 - rpn_cls: 2.1216 - rpn_regr: 0.3420 - final_cls: 0.9685 - final_regr: 0.4191Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 157/1000 [===>..........................] - ETA: 12:51 - rpn_cls: 2.1237 - rpn_regr: 0.3423 - final_cls: 0.9684 - final_regr: 0.4190Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 172/1000 [====>.........................] - ETA: 12:35 - rpn_cls: 2.1358 - rpn_regr: 0.3437 - final_cls: 0.9678 - final_regr: 0.4187Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 182/1000 [====>.........................] - ETA: 12:21 - rpn_cls: 2.1418 - rpn_regr: 0.3444 - final_cls: 0.9677 - final_regr: 0.4185Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 211/1000 [=====>........................] - ETA: 11:45 - rpn_cls: 2.1538 - rpn_regr: 0.3449 - final_cls: 0.9672 - final_regr: 0.4172Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 215/1000 [=====>........................] - ETA: 11:42 - rpn_cls: 2.1546 - rpn_regr: 0.3448 - final_cls: 0.9671 - final_regr: 0.4170Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 230/1000 [=====>........................] - ETA: 11:24 - rpn_cls: 2.1558 - rpn_regr: 0.3441 - final_cls: 0.9667 - final_regr: 0.4164Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 256/1000 [======>.......................] - ETA: 10:57 - rpn_cls: 2.1604 - rpn_regr: 0.3429 - final_cls: 0.9660 - final_regr: 0.4156Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 263/1000 [======>.......................] - ETA: 10:51 - rpn_cls: 2.1620 - rpn_regr: 0.3427 - final_cls: 0.9658 - final_regr: 0.4154Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 268/1000 [=======>......................] - ETA: 10:49 - rpn_cls: 2.1634 - rpn_regr: 0.3425 - final_cls: 0.9657 - final_regr: 0.4153Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 270/1000 [=======>......................] - ETA: 10:49 - rpn_cls: 2.1640 - rpn_regr: 0.3424 - final_cls: 0.9656 - final_regr: 0.4152Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 284/1000 [=======>......................] - ETA: 10:36 - rpn_cls: 2.1676 - rpn_regr: 0.3420 - final_cls: 0.9653 - final_regr: 0.4148Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 298/1000 [=======>......................] - ETA: 10:22 - rpn_cls: 2.1705 - rpn_regr: 0.3417 - final_cls: 0.9649 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 334/1000 [=========>....................] - ETA: 9:40 - rpn_cls: 2.1729 - rpn_regr: 0.3407 - final_cls: 0.9636 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 336/1000 [=========>....................] - ETA: 9:40 - rpn_cls: 2.1729 - rpn_regr: 0.3407 - final_cls: 0.9635 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 351/1000 [=========>....................] - ETA: 9:26 - rpn_cls: 2.1739 - rpn_regr: 0.3405 - final_cls: 0.9631 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 368/1000 [==========>...................] - ETA: 9:09 - rpn_cls: 2.1747 - rpn_regr: 0.3402 - final_cls: 0.9628 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 385/1000 [==========>...................] - ETA: 8:54 - rpn_cls: 2.1751 - rpn_regr: 0.3400 - final_cls: 0.9626 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 411/1000 [===========>..................] - ETA: 8:30 - rpn_cls: 2.1774 - rpn_regr: 0.3402 - final_cls: 0.9625 - final_regr: 0.4141Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 414/1000 [===========>..................] - ETA: 8:28 - rpn_cls: 2.1776 - rpn_regr: 0.3403 - final_cls: 0.9625 - final_regr: 0.4141Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 432/1000 [===========>..................] - ETA: 8:13 - rpn_cls: 2.1802 - rpn_regr: 0.3406 - final_cls: 0.9624 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 441/1000 [============>.................] - ETA: 8:07 - rpn_cls: 2.1822 - rpn_regr: 0.3408 - final_cls: 0.9624 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 451/1000 [============>.................] - ETA: 7:58 - rpn_cls: 2.1849 - rpn_regr: 0.3410 - final_cls: 0.9624 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 453/1000 [============>.................] - ETA: 7:57 - rpn_cls: 2.1854 - rpn_regr: 0.3411 - final_cls: 0.9624 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 472/1000 [=============>................] - ETA: 7:40 - rpn_cls: 2.1894 - rpn_regr: 0.3414 - final_cls: 0.9625 - final_regr: 0.4143Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 481/1000 [=============>................] - ETA: 7:32 - rpn_cls: 2.1906 - rpn_regr: 0.3415 - final_cls: 0.9626 - final_regr: 0.4143Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 482/1000 [=============>................] - ETA: 7:32 - rpn_cls: 2.1907 - rpn_regr: 0.3415 - final_cls: 0.9626 - final_regr: 0.4143Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 490/1000 [=============>................] - ETA: 7:26 - rpn_cls: 2.1916 - rpn_regr: 0.3416 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 506/1000 [==============>...............] - ETA: 7:10 - rpn_cls: 2.1925 - rpn_regr: 0.3416 - final_cls: 0.9626 - final_regr: 0.4145Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 508/1000 [==============>...............] - ETA: 7:09 - rpn_cls: 2.1926 - rpn_regr: 0.3416 - final_cls: 0.9626 - final_regr: 0.4145Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 523/1000 [==============>...............] - ETA: 6:55 - rpn_cls: 2.1931 - rpn_regr: 0.3415 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 535/1000 [===============>..............] - ETA: 6:44 - rpn_cls: 2.1936 - rpn_regr: 0.3414 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 562/1000 [===============>..............] - ETA: 6:20 - rpn_cls: 2.1949 - rpn_regr: 0.3412 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 564/1000 [===============>..............] - ETA: 6:18 - rpn_cls: 2.1950 - rpn_regr: 0.3412 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 570/1000 [================>.............] - ETA: 6:14 - rpn_cls: 2.1953 - rpn_regr: 0.3411 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 580/1000 [================>.............] - ETA: 6:06 - rpn_cls: 2.1960 - rpn_regr: 0.3410 - final_cls: 0.9626 - final_regr: 0.4146Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 596/1000 [================>.............] - ETA: 5:51 - rpn_cls: 2.1976 - rpn_regr: 0.3408 - final_cls: 0.9626 - final_regr: 0.4145Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 602/1000 [=================>............] - ETA: 5:46 - rpn_cls: 2.1982 - rpn_regr: 0.3408 - final_cls: 0.9626 - final_regr: 0.4145Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 624/1000 [=================>............] - ETA: 5:25 - rpn_cls: 2.2006 - rpn_regr: 0.3405 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 628/1000 [=================>............] - ETA: 5:22 - rpn_cls: 2.2010 - rpn_regr: 0.3404 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 636/1000 [==================>...........] - ETA: 5:15 - rpn_cls: 2.2017 - rpn_regr: 0.3404 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 638/1000 [==================>...........] - ETA: 5:14 - rpn_cls: 2.2018 - rpn_regr: 0.3403 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 641/1000 [==================>...........] - ETA: 5:12 - rpn_cls: 2.2021 - rpn_regr: 0.3403 - final_cls: 0.9626 - final_regr: 0.4144Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 645/1000 [==================>...........] - ETA: 5:09 - rpn_cls: 2.2023 - rpn_regr: 0.3403 - final_cls: 0.9626 - final_regr: 0.4143Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 672/1000 [===================>..........] - ETA: 4:45 - rpn_cls: 2.2040 - rpn_regr: 0.3399 - final_cls: 0.9625 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 673/1000 [===================>..........] - ETA: 4:44 - rpn_cls: 2.2041 - rpn_regr: 0.3399 - final_cls: 0.9625 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 685/1000 [===================>..........] - ETA: 4:34 - rpn_cls: 2.2050 - rpn_regr: 0.3398 - final_cls: 0.9625 - final_regr: 0.4142Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 698/1000 [===================>..........] - ETA: 4:23 - rpn_cls: 2.2064 - rpn_regr: 0.3398 - final_cls: 0.9624 - final_regr: 0.4141Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 701/1000 [====================>.........] - ETA: 4:20 - rpn_cls: 2.2067 - rpn_regr: 0.3398 - final_cls: 0.9624 - final_regr: 0.4141Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 719/1000 [====================>.........] - ETA: 4:04 - rpn_cls: 2.2091 - rpn_regr: 0.3397 - final_cls: 0.9622 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 728/1000 [====================>.........] - ETA: 3:56 - rpn_cls: 2.2106 - rpn_regr: 0.3397 - final_cls: 0.9621 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 742/1000 [=====================>........] - ETA: 3:44 - rpn_cls: 2.2129 - rpn_regr: 0.3397 - final_cls: 0.9621 - final_regr: 0.4140Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 757/1000 [=====================>........] - ETA: 3:31 - rpn_cls: 2.2151 - rpn_regr: 0.3396 - final_cls: 0.9620 - final_regr: 0.4139Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 773/1000 [======================>.......] - ETA: 3:17 - rpn_cls: 2.2172 - rpn_regr: 0.3396 - final_cls: 0.9619 - final_regr: 0.4139Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 783/1000 [======================>.......] - ETA: 3:08 - rpn_cls: 2.2185 - rpn_regr: 0.3395 - final_cls: 0.9619 - final_regr: 0.4139Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 784/1000 [======================>.......] - ETA: 3:07 - rpn_cls: 2.2187 - rpn_regr: 0.3395 - final_cls: 0.9619 - final_regr: 0.4139Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 794/1000 [======================>.......] - ETA: 2:59 - rpn_cls: 2.2200 - rpn_regr: 0.3395 - final_cls: 0.9618 - final_regr: 0.4139Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 797/1000 [======================>.......] - ETA: 2:56 - rpn_cls: 2.2204 - rpn_regr: 0.3395 - final_cls: 0.9618 - final_regr: 0.4138Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 840/1000 [========================>.....] - ETA: 2:18 - rpn_cls: 2.2257 - rpn_regr: 0.3391 - final_cls: 0.9616 - final_regr: 0.4137Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 846/1000 [========================>.....] - ETA: 2:13 - rpn_cls: 2.2264 - rpn_regr: 0.3391 - final_cls: 0.9616 - final_regr: 0.4137Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 855/1000 [========================>.....] - ETA: 2:05 - rpn_cls: 2.2275 - rpn_regr: 0.3390 - final_cls: 0.9615 - final_regr: 0.4137Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 884/1000 [=========================>....] - ETA: 1:40 - rpn_cls: 2.2315 - rpn_regr: 0.3388 - final_cls: 0.9614 - final_regr: 0.4135Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 918/1000 [==========================>...] - ETA: 1:10 - rpn_cls: 2.2356 - rpn_regr: 0.3386 - final_cls: 0.9611 - final_regr: 0.4134Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 919/1000 [==========================>...] - ETA: 1:09 - rpn_cls: 2.2357 - rpn_regr: 0.3386 - final_cls: 0.9611 - final_regr: 0.4134Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 955/1000 [===========================>..] - ETA: 38s - rpn_cls: 2.2397 - rpn_regr: 0.3384 - final_cls: 0.9609 - final_regr: 0.4134Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 970/1000 [============================>.] - ETA: 25s - rpn_cls: 2.2412 - rpn_regr: 0.3384 - final_cls: 0.9608 - final_regr: 0.4134Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 998/1000 [============================>.] - ETA: 1s - rpn_cls: 2.2438 - rpn_regr: 0.3383 - final_cls: 0.9606 - final_regr: 0.4133Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 860s 861ms/step - rpn_cls: 2.2440 - rpn_regr: 0.3383 - final_cls: 0.9606 - final_regr: 0.4133\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.390423572744014\n",
            "Classifier accuracy for bounding boxes from RPN: 0.55075\n",
            "Loss RPN classifier: 2.336463462337539\n",
            "Loss RPN regression: 0.3355526297418401\n",
            "Loss Detector classifier: 0.9537065534591674\n",
            "Loss Detector regression: 0.41120719480142\n",
            "Total loss: 4.036929840339966\n",
            "Elapsed time: 862.908303976059\n",
            "Total loss decreased from 4.219649989471123 to 4.036929840339966, saving weights\n",
            "Epoch 11/40\n",
            "   1/1000 [..............................] - ETA: 16:18 - rpn_cls: 1.8706 - rpn_regr: 0.4354 - final_cls: 0.7125 - final_regr: 0.2761Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  19/1000 [..............................] - ETA: 15:10 - rpn_cls: 3.1435 - rpn_regr: 0.3387 - final_cls: 0.8638 - final_regr: 0.3403Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  28/1000 [..............................] - ETA: 15:19 - rpn_cls: 3.2329 - rpn_regr: 0.3481 - final_cls: 0.8698 - final_regr: 0.3521Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  41/1000 [>.............................] - ETA: 15:10 - rpn_cls: 3.1352 - rpn_regr: 0.3481 - final_cls: 0.8878 - final_regr: 0.3610Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  56/1000 [>.............................] - ETA: 14:41 - rpn_cls: 3.0035 - rpn_regr: 0.3456 - final_cls: 0.9056 - final_regr: 0.3690Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  66/1000 [>.............................] - ETA: 14:18 - rpn_cls: 2.9317 - rpn_regr: 0.3443 - final_cls: 0.9129 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  69/1000 [=>............................] - ETA: 14:24 - rpn_cls: 2.9090 - rpn_regr: 0.3437 - final_cls: 0.9142 - final_regr: 0.3746Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  79/1000 [=>............................] - ETA: 14:22 - rpn_cls: 2.8329 - rpn_regr: 0.3421 - final_cls: 0.9190 - final_regr: 0.3781Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  82/1000 [=>............................] - ETA: 14:24 - rpn_cls: 2.8134 - rpn_regr: 0.3414 - final_cls: 0.9200 - final_regr: 0.3793Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  92/1000 [=>............................] - ETA: 14:16 - rpn_cls: 2.7516 - rpn_regr: 0.3383 - final_cls: 0.9216 - final_regr: 0.3827Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  96/1000 [=>............................] - ETA: 14:12 - rpn_cls: 2.7346 - rpn_regr: 0.3369 - final_cls: 0.9223 - final_regr: 0.3837Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 107/1000 [==>...........................] - ETA: 13:56 - rpn_cls: 2.6907 - rpn_regr: 0.3329 - final_cls: 0.9229 - final_regr: 0.3857Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 110/1000 [==>...........................] - ETA: 13:56 - rpn_cls: 2.6798 - rpn_regr: 0.3318 - final_cls: 0.9229 - final_regr: 0.3861Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 122/1000 [==>...........................] - ETA: 13:40 - rpn_cls: 2.6448 - rpn_regr: 0.3281 - final_cls: 0.9224 - final_regr: 0.3870Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 135/1000 [===>..........................] - ETA: 13:33 - rpn_cls: 2.6123 - rpn_regr: 0.3251 - final_cls: 0.9218 - final_regr: 0.3875Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 137/1000 [===>..........................] - ETA: 13:35 - rpn_cls: 2.6083 - rpn_regr: 0.3247 - final_cls: 0.9216 - final_regr: 0.3874Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 144/1000 [===>..........................] - ETA: 13:26 - rpn_cls: 2.5945 - rpn_regr: 0.3235 - final_cls: 0.9214 - final_regr: 0.3876Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 147/1000 [===>..........................] - ETA: 13:24 - rpn_cls: 2.5893 - rpn_regr: 0.3230 - final_cls: 0.9214 - final_regr: 0.3878Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 149/1000 [===>..........................] - ETA: 13:25 - rpn_cls: 2.5860 - rpn_regr: 0.3228 - final_cls: 0.9213 - final_regr: 0.3879Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 155/1000 [===>..........................] - ETA: 13:20 - rpn_cls: 2.5782 - rpn_regr: 0.3223 - final_cls: 0.9210 - final_regr: 0.3882Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 156/1000 [===>..........................] - ETA: 13:24 - rpn_cls: 2.5770 - rpn_regr: 0.3222 - final_cls: 0.9209 - final_regr: 0.3882Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 188/1000 [====>.........................] - ETA: 12:33 - rpn_cls: 2.5502 - rpn_regr: 0.3191 - final_cls: 0.9206 - final_regr: 0.3894Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 215/1000 [=====>........................] - ETA: 11:54 - rpn_cls: 2.5285 - rpn_regr: 0.3167 - final_cls: 0.9198 - final_regr: 0.3898Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 220/1000 [=====>........................] - ETA: 11:49 - rpn_cls: 2.5245 - rpn_regr: 0.3164 - final_cls: 0.9197 - final_regr: 0.3899Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 229/1000 [=====>........................] - ETA: 11:42 - rpn_cls: 2.5171 - rpn_regr: 0.3159 - final_cls: 0.9197 - final_regr: 0.3901Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 233/1000 [=====>........................] - ETA: 11:38 - rpn_cls: 2.5138 - rpn_regr: 0.3156 - final_cls: 0.9197 - final_regr: 0.3902Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 240/1000 [======>.......................] - ETA: 11:30 - rpn_cls: 2.5072 - rpn_regr: 0.3152 - final_cls: 0.9195 - final_regr: 0.3903Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 244/1000 [======>.......................] - ETA: 11:27 - rpn_cls: 2.5032 - rpn_regr: 0.3150 - final_cls: 0.9194 - final_regr: 0.3904Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 259/1000 [======>.......................] - ETA: 11:11 - rpn_cls: 2.4898 - rpn_regr: 0.3142 - final_cls: 0.9192 - final_regr: 0.3908Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 273/1000 [=======>......................] - ETA: 10:57 - rpn_cls: 2.4789 - rpn_regr: 0.3139 - final_cls: 0.9190 - final_regr: 0.3912Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 286/1000 [=======>......................] - ETA: 10:43 - rpn_cls: 2.4710 - rpn_regr: 0.3140 - final_cls: 0.9190 - final_regr: 0.3914Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 289/1000 [=======>......................] - ETA: 10:41 - rpn_cls: 2.4692 - rpn_regr: 0.3140 - final_cls: 0.9190 - final_regr: 0.3915Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 305/1000 [========>.....................] - ETA: 10:28 - rpn_cls: 2.4616 - rpn_regr: 0.3142 - final_cls: 0.9191 - final_regr: 0.3916Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 306/1000 [========>.....................] - ETA: 10:28 - rpn_cls: 2.4613 - rpn_regr: 0.3142 - final_cls: 0.9191 - final_regr: 0.3916Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 309/1000 [========>.....................] - ETA: 10:26 - rpn_cls: 2.4602 - rpn_regr: 0.3143 - final_cls: 0.9191 - final_regr: 0.3916Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 314/1000 [========>.....................] - ETA: 10:24 - rpn_cls: 2.4583 - rpn_regr: 0.3144 - final_cls: 0.9192 - final_regr: 0.3917Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 324/1000 [========>.....................] - ETA: 10:14 - rpn_cls: 2.4544 - rpn_regr: 0.3146 - final_cls: 0.9192 - final_regr: 0.3918Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 326/1000 [========>.....................] - ETA: 10:15 - rpn_cls: 2.4536 - rpn_regr: 0.3146 - final_cls: 0.9192 - final_regr: 0.3918Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 341/1000 [=========>....................] - ETA: 10:00 - rpn_cls: 2.4473 - rpn_regr: 0.3150 - final_cls: 0.9194 - final_regr: 0.3919Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 366/1000 [=========>....................] - ETA: 9:32 - rpn_cls: 2.4390 - rpn_regr: 0.3153 - final_cls: 0.9198 - final_regr: 0.3919Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 380/1000 [==========>...................] - ETA: 9:18 - rpn_cls: 2.4339 - rpn_regr: 0.3153 - final_cls: 0.9201 - final_regr: 0.3920Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 384/1000 [==========>...................] - ETA: 9:15 - rpn_cls: 2.4325 - rpn_regr: 0.3153 - final_cls: 0.9202 - final_regr: 0.3920Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 399/1000 [==========>...................] - ETA: 8:59 - rpn_cls: 2.4271 - rpn_regr: 0.3152 - final_cls: 0.9204 - final_regr: 0.3921Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 425/1000 [===========>..................] - ETA: 8:35 - rpn_cls: 2.4188 - rpn_regr: 0.3149 - final_cls: 0.9208 - final_regr: 0.3923Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 431/1000 [===========>..................] - ETA: 8:30 - rpn_cls: 2.4170 - rpn_regr: 0.3149 - final_cls: 0.9209 - final_regr: 0.3923Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 436/1000 [============>.................] - ETA: 8:27 - rpn_cls: 2.4155 - rpn_regr: 0.3148 - final_cls: 0.9210 - final_regr: 0.3923Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 501/1000 [==============>...............] - ETA: 7:22 - rpn_cls: 2.4025 - rpn_regr: 0.3148 - final_cls: 0.9216 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 504/1000 [==============>...............] - ETA: 7:20 - rpn_cls: 2.4022 - rpn_regr: 0.3148 - final_cls: 0.9216 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 515/1000 [==============>...............] - ETA: 7:12 - rpn_cls: 2.4006 - rpn_regr: 0.3148 - final_cls: 0.9218 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 523/1000 [==============>...............] - ETA: 7:04 - rpn_cls: 2.3995 - rpn_regr: 0.3148 - final_cls: 0.9220 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 535/1000 [===============>..............] - ETA: 6:53 - rpn_cls: 2.3977 - rpn_regr: 0.3148 - final_cls: 0.9222 - final_regr: 0.3927Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 552/1000 [===============>..............] - ETA: 6:37 - rpn_cls: 2.3952 - rpn_regr: 0.3149 - final_cls: 0.9224 - final_regr: 0.3927Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 564/1000 [===============>..............] - ETA: 6:26 - rpn_cls: 2.3933 - rpn_regr: 0.3150 - final_cls: 0.9226 - final_regr: 0.3928Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 577/1000 [================>.............] - ETA: 6:14 - rpn_cls: 2.3913 - rpn_regr: 0.3152 - final_cls: 0.9228 - final_regr: 0.3928Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 599/1000 [================>.............] - ETA: 5:55 - rpn_cls: 2.3884 - rpn_regr: 0.3157 - final_cls: 0.9231 - final_regr: 0.3929Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 605/1000 [=================>............] - ETA: 5:50 - rpn_cls: 2.3878 - rpn_regr: 0.3158 - final_cls: 0.9232 - final_regr: 0.3929Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 607/1000 [=================>............] - ETA: 5:49 - rpn_cls: 2.3875 - rpn_regr: 0.3158 - final_cls: 0.9232 - final_regr: 0.3929Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 621/1000 [=================>............] - ETA: 5:36 - rpn_cls: 2.3858 - rpn_regr: 0.3161 - final_cls: 0.9234 - final_regr: 0.3930Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 629/1000 [=================>............] - ETA: 5:29 - rpn_cls: 2.3847 - rpn_regr: 0.3163 - final_cls: 0.9235 - final_regr: 0.3931Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 635/1000 [==================>...........] - ETA: 5:24 - rpn_cls: 2.3840 - rpn_regr: 0.3164 - final_cls: 0.9236 - final_regr: 0.3931Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 651/1000 [==================>...........] - ETA: 5:09 - rpn_cls: 2.3822 - rpn_regr: 0.3166 - final_cls: 0.9238 - final_regr: 0.3931Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 659/1000 [==================>...........] - ETA: 5:02 - rpn_cls: 2.3814 - rpn_regr: 0.3167 - final_cls: 0.9239 - final_regr: 0.3931Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 662/1000 [==================>...........] - ETA: 5:00 - rpn_cls: 2.3812 - rpn_regr: 0.3168 - final_cls: 0.9240 - final_regr: 0.3931Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 720/1000 [====================>.........] - ETA: 4:06 - rpn_cls: 2.3768 - rpn_regr: 0.3173 - final_cls: 0.9245 - final_regr: 0.3932Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 736/1000 [=====================>........] - ETA: 3:52 - rpn_cls: 2.3758 - rpn_regr: 0.3174 - final_cls: 0.9247 - final_regr: 0.3932Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 752/1000 [=====================>........] - ETA: 3:38 - rpn_cls: 2.3750 - rpn_regr: 0.3175 - final_cls: 0.9248 - final_regr: 0.3933Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 767/1000 [======================>.......] - ETA: 3:24 - rpn_cls: 2.3742 - rpn_regr: 0.3177 - final_cls: 0.9249 - final_regr: 0.3933Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 768/1000 [======================>.......] - ETA: 3:23 - rpn_cls: 2.3742 - rpn_regr: 0.3177 - final_cls: 0.9249 - final_regr: 0.3933Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 781/1000 [======================>.......] - ETA: 3:12 - rpn_cls: 2.3735 - rpn_regr: 0.3177 - final_cls: 0.9250 - final_regr: 0.3933Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 796/1000 [======================>.......] - ETA: 2:58 - rpn_cls: 2.3726 - rpn_regr: 0.3179 - final_cls: 0.9251 - final_regr: 0.3934Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 799/1000 [======================>.......] - ETA: 2:56 - rpn_cls: 2.3723 - rpn_regr: 0.3179 - final_cls: 0.9252 - final_regr: 0.3934Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 808/1000 [=======================>......] - ETA: 2:48 - rpn_cls: 2.3717 - rpn_regr: 0.3179 - final_cls: 0.9252 - final_regr: 0.3935Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 827/1000 [=======================>......] - ETA: 2:31 - rpn_cls: 2.3702 - rpn_regr: 0.3181 - final_cls: 0.9254 - final_regr: 0.3936Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 831/1000 [=======================>......] - ETA: 2:27 - rpn_cls: 2.3698 - rpn_regr: 0.3181 - final_cls: 0.9254 - final_regr: 0.3936Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 848/1000 [========================>.....] - ETA: 2:12 - rpn_cls: 2.3684 - rpn_regr: 0.3182 - final_cls: 0.9256 - final_regr: 0.3937Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 874/1000 [=========================>....] - ETA: 1:49 - rpn_cls: 2.3664 - rpn_regr: 0.3185 - final_cls: 0.9258 - final_regr: 0.3938Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 894/1000 [=========================>....] - ETA: 1:32 - rpn_cls: 2.3653 - rpn_regr: 0.3187 - final_cls: 0.9258 - final_regr: 0.3939Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 895/1000 [=========================>....] - ETA: 1:31 - rpn_cls: 2.3652 - rpn_regr: 0.3187 - final_cls: 0.9259 - final_regr: 0.3939Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 899/1000 [=========================>....] - ETA: 1:28 - rpn_cls: 2.3650 - rpn_regr: 0.3188 - final_cls: 0.9259 - final_regr: 0.3939Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 917/1000 [==========================>...] - ETA: 1:12 - rpn_cls: 2.3639 - rpn_regr: 0.3190 - final_cls: 0.9260 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 932/1000 [==========================>...] - ETA: 59s - rpn_cls: 2.3630 - rpn_regr: 0.3191 - final_cls: 0.9261 - final_regr: 0.3941 Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 936/1000 [===========================>..] - ETA: 55s - rpn_cls: 2.3627 - rpn_regr: 0.3192 - final_cls: 0.9262 - final_regr: 0.3941Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 941/1000 [===========================>..] - ETA: 51s - rpn_cls: 2.3624 - rpn_regr: 0.3192 - final_cls: 0.9262 - final_regr: 0.3942Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 970/1000 [============================>.] - ETA: 26s - rpn_cls: 2.3602 - rpn_regr: 0.3195 - final_cls: 0.9264 - final_regr: 0.3943Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 869s 869ms/step - rpn_cls: 2.3581 - rpn_regr: 0.3197 - final_cls: 0.9266 - final_regr: 0.3944\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.559926806953339\n",
            "Classifier accuracy for bounding boxes from RPN: 0.5545\n",
            "Loss RPN classifier: 2.2935821444237794\n",
            "Loss RPN regression: 0.32438344323262575\n",
            "Loss Detector classifier: 0.9317936573326587\n",
            "Loss Detector regression: 0.398914384521544\n",
            "Total loss: 3.948673629510608\n",
            "Elapsed time: 871.9082546234131\n",
            "Total loss decreased from 4.036929840339966 to 3.948673629510608, saving weights\n",
            "Epoch 12/40\n",
            "   3/1000 [..............................] - ETA: 14:05 - rpn_cls: 0.6173 - rpn_regr: 0.3261 - final_cls: 0.8767 - final_regr: 0.2801Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  23/1000 [..............................] - ETA: 15:28 - rpn_cls: 1.9304 - rpn_regr: 0.3320 - final_cls: 0.8750 - final_regr: 0.3857Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  28/1000 [..............................] - ETA: 15:46 - rpn_cls: 1.9566 - rpn_regr: 0.3253 - final_cls: 0.8880 - final_regr: 0.3867Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  30/1000 [..............................] - ETA: 15:58 - rpn_cls: 1.9582 - rpn_regr: 0.3234 - final_cls: 0.8934 - final_regr: 0.3864Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  46/1000 [>.............................] - ETA: 15:25 - rpn_cls: 1.9478 - rpn_regr: 0.3306 - final_cls: 0.9190 - final_regr: 0.3851Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  58/1000 [>.............................] - ETA: 14:48 - rpn_cls: 1.9963 - rpn_regr: 0.3308 - final_cls: 0.9268 - final_regr: 0.3885Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  92/1000 [=>............................] - ETA: 13:27 - rpn_cls: 2.0838 - rpn_regr: 0.3280 - final_cls: 0.9354 - final_regr: 0.3927Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  95/1000 [=>............................] - ETA: 13:31 - rpn_cls: 2.0879 - rpn_regr: 0.3279 - final_cls: 0.9362 - final_regr: 0.3932Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 140/1000 [===>..........................] - ETA: 12:32 - rpn_cls: 2.1086 - rpn_regr: 0.3250 - final_cls: 0.9419 - final_regr: 0.3991Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 143/1000 [===>..........................] - ETA: 12:35 - rpn_cls: 2.1090 - rpn_regr: 0.3251 - final_cls: 0.9420 - final_regr: 0.3993Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 144/1000 [===>..........................] - ETA: 12:41 - rpn_cls: 2.1089 - rpn_regr: 0.3251 - final_cls: 0.9420 - final_regr: 0.3994Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 170/1000 [====>.........................] - ETA: 12:11 - rpn_cls: 2.1122 - rpn_regr: 0.3261 - final_cls: 0.9426 - final_regr: 0.4005Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 215/1000 [=====>........................] - ETA: 11:24 - rpn_cls: 2.1237 - rpn_regr: 0.3309 - final_cls: 0.9404 - final_regr: 0.4019Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 230/1000 [=====>........................] - ETA: 11:10 - rpn_cls: 2.1242 - rpn_regr: 0.3322 - final_cls: 0.9400 - final_regr: 0.4020Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 240/1000 [======>.......................] - ETA: 10:59 - rpn_cls: 2.1249 - rpn_regr: 0.3329 - final_cls: 0.9397 - final_regr: 0.4019Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 243/1000 [======>.......................] - ETA: 10:58 - rpn_cls: 2.1247 - rpn_regr: 0.3331 - final_cls: 0.9396 - final_regr: 0.4018Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 244/1000 [======>.......................] - ETA: 10:59 - rpn_cls: 2.1246 - rpn_regr: 0.3331 - final_cls: 0.9396 - final_regr: 0.4018Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 284/1000 [=======>......................] - ETA: 10:17 - rpn_cls: 2.1196 - rpn_regr: 0.3336 - final_cls: 0.9381 - final_regr: 0.4010Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 300/1000 [========>.....................] - ETA: 10:01 - rpn_cls: 2.1183 - rpn_regr: 0.3332 - final_cls: 0.9374 - final_regr: 0.4007Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 330/1000 [========>.....................] - ETA: 9:32 - rpn_cls: 2.1197 - rpn_regr: 0.3328 - final_cls: 0.9363 - final_regr: 0.4000Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 346/1000 [=========>....................] - ETA: 9:18 - rpn_cls: 2.1206 - rpn_regr: 0.3327 - final_cls: 0.9359 - final_regr: 0.3997Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 395/1000 [==========>...................] - ETA: 8:30 - rpn_cls: 2.1280 - rpn_regr: 0.3323 - final_cls: 0.9349 - final_regr: 0.3987Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 396/1000 [==========>...................] - ETA: 8:30 - rpn_cls: 2.1282 - rpn_regr: 0.3323 - final_cls: 0.9349 - final_regr: 0.3987Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 408/1000 [===========>..................] - ETA: 8:21 - rpn_cls: 2.1304 - rpn_regr: 0.3322 - final_cls: 0.9348 - final_regr: 0.3986Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 445/1000 [============>.................] - ETA: 7:46 - rpn_cls: 2.1360 - rpn_regr: 0.3321 - final_cls: 0.9344 - final_regr: 0.3979Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 446/1000 [============>.................] - ETA: 7:47 - rpn_cls: 2.1362 - rpn_regr: 0.3321 - final_cls: 0.9344 - final_regr: 0.3979Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 458/1000 [============>.................] - ETA: 7:37 - rpn_cls: 2.1391 - rpn_regr: 0.3322 - final_cls: 0.9342 - final_regr: 0.3976Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 492/1000 [=============>................] - ETA: 7:07 - rpn_cls: 2.1477 - rpn_regr: 0.3327 - final_cls: 0.9340 - final_regr: 0.3971Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 493/1000 [=============>................] - ETA: 7:07 - rpn_cls: 2.1480 - rpn_regr: 0.3328 - final_cls: 0.9340 - final_regr: 0.3971Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 499/1000 [=============>................] - ETA: 7:02 - rpn_cls: 2.1500 - rpn_regr: 0.3329 - final_cls: 0.9339 - final_regr: 0.3970Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 501/1000 [==============>...............] - ETA: 7:01 - rpn_cls: 2.1506 - rpn_regr: 0.3329 - final_cls: 0.9339 - final_regr: 0.3970Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 514/1000 [==============>...............] - ETA: 6:51 - rpn_cls: 2.1545 - rpn_regr: 0.3332 - final_cls: 0.9340 - final_regr: 0.3968Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 533/1000 [==============>...............] - ETA: 6:36 - rpn_cls: 2.1596 - rpn_regr: 0.3335 - final_cls: 0.9341 - final_regr: 0.3966Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 538/1000 [===============>..............] - ETA: 6:32 - rpn_cls: 2.1609 - rpn_regr: 0.3336 - final_cls: 0.9342 - final_regr: 0.3965Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 552/1000 [===============>..............] - ETA: 6:19 - rpn_cls: 2.1641 - rpn_regr: 0.3337 - final_cls: 0.9343 - final_regr: 0.3964Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 555/1000 [===============>..............] - ETA: 6:17 - rpn_cls: 2.1648 - rpn_regr: 0.3337 - final_cls: 0.9343 - final_regr: 0.3963Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 565/1000 [===============>..............] - ETA: 6:09 - rpn_cls: 2.1668 - rpn_regr: 0.3338 - final_cls: 0.9343 - final_regr: 0.3962Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 581/1000 [================>.............] - ETA: 5:54 - rpn_cls: 2.1693 - rpn_regr: 0.3338 - final_cls: 0.9343 - final_regr: 0.3960Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 597/1000 [================>.............] - ETA: 5:40 - rpn_cls: 2.1713 - rpn_regr: 0.3337 - final_cls: 0.9343 - final_regr: 0.3958Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 618/1000 [=================>............] - ETA: 5:23 - rpn_cls: 2.1732 - rpn_regr: 0.3336 - final_cls: 0.9341 - final_regr: 0.3955Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 627/1000 [=================>............] - ETA: 5:15 - rpn_cls: 2.1739 - rpn_regr: 0.3335 - final_cls: 0.9341 - final_regr: 0.3954Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 633/1000 [=================>............] - ETA: 5:10 - rpn_cls: 2.1744 - rpn_regr: 0.3335 - final_cls: 0.9341 - final_regr: 0.3954Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 642/1000 [==================>...........] - ETA: 5:03 - rpn_cls: 2.1753 - rpn_regr: 0.3334 - final_cls: 0.9340 - final_regr: 0.3953Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 657/1000 [==================>...........] - ETA: 4:50 - rpn_cls: 2.1768 - rpn_regr: 0.3333 - final_cls: 0.9340 - final_regr: 0.3952Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 658/1000 [==================>...........] - ETA: 4:50 - rpn_cls: 2.1769 - rpn_regr: 0.3333 - final_cls: 0.9340 - final_regr: 0.3952Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 687/1000 [===================>..........] - ETA: 4:24 - rpn_cls: 2.1782 - rpn_regr: 0.3331 - final_cls: 0.9339 - final_regr: 0.3949Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 690/1000 [===================>..........] - ETA: 4:22 - rpn_cls: 2.1783 - rpn_regr: 0.3331 - final_cls: 0.9339 - final_regr: 0.3949Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 699/1000 [===================>..........] - ETA: 4:15 - rpn_cls: 2.1787 - rpn_regr: 0.3330 - final_cls: 0.9338 - final_regr: 0.3948Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 705/1000 [====================>.........] - ETA: 4:10 - rpn_cls: 2.1789 - rpn_regr: 0.3329 - final_cls: 0.9338 - final_regr: 0.3947Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 709/1000 [====================>.........] - ETA: 4:07 - rpn_cls: 2.1791 - rpn_regr: 0.3329 - final_cls: 0.9338 - final_regr: 0.3947Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 737/1000 [=====================>........] - ETA: 3:42 - rpn_cls: 2.1801 - rpn_regr: 0.3327 - final_cls: 0.9336 - final_regr: 0.3945Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 738/1000 [=====================>........] - ETA: 3:41 - rpn_cls: 2.1801 - rpn_regr: 0.3327 - final_cls: 0.9336 - final_regr: 0.3945Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 768/1000 [======================>.......] - ETA: 3:16 - rpn_cls: 2.1820 - rpn_regr: 0.3326 - final_cls: 0.9335 - final_regr: 0.3943Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 786/1000 [======================>.......] - ETA: 3:01 - rpn_cls: 2.1832 - rpn_regr: 0.3326 - final_cls: 0.9335 - final_regr: 0.3942Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 789/1000 [======================>.......] - ETA: 2:58 - rpn_cls: 2.1834 - rpn_regr: 0.3326 - final_cls: 0.9335 - final_regr: 0.3942Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 809/1000 [=======================>......] - ETA: 2:41 - rpn_cls: 2.1848 - rpn_regr: 0.3326 - final_cls: 0.9335 - final_regr: 0.3942Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 824/1000 [=======================>......] - ETA: 2:29 - rpn_cls: 2.1858 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3941Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 828/1000 [=======================>......] - ETA: 2:25 - rpn_cls: 2.1861 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3941Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 833/1000 [=======================>......] - ETA: 2:21 - rpn_cls: 2.1865 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3941Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 836/1000 [========================>.....] - ETA: 2:19 - rpn_cls: 2.1866 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3941Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 850/1000 [========================>.....] - ETA: 2:07 - rpn_cls: 2.1874 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 857/1000 [========================>.....] - ETA: 2:01 - rpn_cls: 2.1877 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 859/1000 [========================>.....] - ETA: 1:59 - rpn_cls: 2.1878 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 863/1000 [========================>.....] - ETA: 1:56 - rpn_cls: 2.1880 - rpn_regr: 0.3325 - final_cls: 0.9335 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 929/1000 [==========================>...] - ETA: 1:00 - rpn_cls: 2.1896 - rpn_regr: 0.3320 - final_cls: 0.9334 - final_regr: 0.3937Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 949/1000 [===========================>..] - ETA: 43s - rpn_cls: 2.1897 - rpn_regr: 0.3319 - final_cls: 0.9334 - final_regr: 0.3936Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 955/1000 [===========================>..] - ETA: 38s - rpn_cls: 2.1897 - rpn_regr: 0.3319 - final_cls: 0.9334 - final_regr: 0.3936Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 984/1000 [============================>.] - ETA: 13s - rpn_cls: 2.1895 - rpn_regr: 0.3317 - final_cls: 0.9334 - final_regr: 0.3935Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 988/1000 [============================>.] - ETA: 10s - rpn_cls: 2.1895 - rpn_regr: 0.3316 - final_cls: 0.9334 - final_regr: 0.3935Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 843s 843ms/step - rpn_cls: 2.1894 - rpn_regr: 0.3316 - final_cls: 0.9334 - final_regr: 0.3934\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.806122448979592\n",
            "Classifier accuracy for bounding boxes from RPN: 0.55875\n",
            "Loss RPN classifier: 2.183746423054565\n",
            "Loss RPN regression: 0.32527628695219757\n",
            "Loss Detector classifier: 0.9322663827240467\n",
            "Loss Detector regression: 0.38962023047730326\n",
            "Total loss: 3.8309093232081124\n",
            "Elapsed time: 845.8625020980835\n",
            "Total loss decreased from 3.948673629510608 to 3.8309093232081124, saving weights\n",
            "Epoch 13/40\n",
            "   3/1000 [..............................] - ETA: 12:04 - rpn_cls: 1.8606 - rpn_regr: 0.5100 - final_cls: 0.9198 - final_regr: 0.6122Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  15/1000 [..............................] - ETA: 12:22 - rpn_cls: 1.4576 - rpn_regr: 0.3515 - final_cls: 0.8740 - final_regr: 0.5109Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  19/1000 [..............................] - ETA: 12:55 - rpn_cls: 1.4584 - rpn_regr: 0.3314 - final_cls: 0.8795 - final_regr: 0.5007Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  36/1000 [>.............................] - ETA: 13:28 - rpn_cls: 1.6206 - rpn_regr: 0.3151 - final_cls: 0.8863 - final_regr: 0.4677Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  62/1000 [>.............................] - ETA: 13:26 - rpn_cls: 1.8345 - rpn_regr: 0.3352 - final_cls: 0.8978 - final_regr: 0.4500Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  65/1000 [>.............................] - ETA: 13:32 - rpn_cls: 1.8485 - rpn_regr: 0.3367 - final_cls: 0.8987 - final_regr: 0.4487Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  72/1000 [=>............................] - ETA: 13:47 - rpn_cls: 1.8737 - rpn_regr: 0.3399 - final_cls: 0.8999 - final_regr: 0.4462Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  80/1000 [=>............................] - ETA: 13:47 - rpn_cls: 1.9018 - rpn_regr: 0.3432 - final_cls: 0.9010 - final_regr: 0.4437Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  81/1000 [=>............................] - ETA: 13:53 - rpn_cls: 1.9050 - rpn_regr: 0.3435 - final_cls: 0.9012 - final_regr: 0.4434Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  87/1000 [=>............................] - ETA: 13:55 - rpn_cls: 1.9213 - rpn_regr: 0.3456 - final_cls: 0.9028 - final_regr: 0.4417Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  89/1000 [=>............................] - ETA: 13:58 - rpn_cls: 1.9261 - rpn_regr: 0.3462 - final_cls: 0.9035 - final_regr: 0.4411Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  99/1000 [=>............................] - ETA: 13:47 - rpn_cls: 1.9522 - rpn_regr: 0.3478 - final_cls: 0.9072 - final_regr: 0.4387Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 102/1000 [==>...........................] - ETA: 13:53 - rpn_cls: 1.9587 - rpn_regr: 0.3480 - final_cls: 0.9082 - final_regr: 0.4380Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 127/1000 [==>...........................] - ETA: 13:15 - rpn_cls: 2.0069 - rpn_regr: 0.3473 - final_cls: 0.9140 - final_regr: 0.4336Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 132/1000 [==>...........................] - ETA: 13:11 - rpn_cls: 2.0141 - rpn_regr: 0.3468 - final_cls: 0.9147 - final_regr: 0.4325Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 144/1000 [===>..........................] - ETA: 13:01 - rpn_cls: 2.0270 - rpn_regr: 0.3452 - final_cls: 0.9165 - final_regr: 0.4299Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 154/1000 [===>..........................] - ETA: 12:51 - rpn_cls: 2.0321 - rpn_regr: 0.3437 - final_cls: 0.9178 - final_regr: 0.4280Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 157/1000 [===>..........................] - ETA: 12:48 - rpn_cls: 2.0334 - rpn_regr: 0.3431 - final_cls: 0.9181 - final_regr: 0.4274Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 199/1000 [====>.........................] - ETA: 12:01 - rpn_cls: 2.0499 - rpn_regr: 0.3358 - final_cls: 0.9207 - final_regr: 0.4198Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 225/1000 [=====>........................] - ETA: 11:31 - rpn_cls: 2.0543 - rpn_regr: 0.3326 - final_cls: 0.9216 - final_regr: 0.4159Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 248/1000 [======>.......................] - ETA: 11:05 - rpn_cls: 2.0613 - rpn_regr: 0.3305 - final_cls: 0.9222 - final_regr: 0.4130Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 279/1000 [=======>......................] - ETA: 10:27 - rpn_cls: 2.0664 - rpn_regr: 0.3283 - final_cls: 0.9229 - final_regr: 0.4097Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 282/1000 [=======>......................] - ETA: 10:26 - rpn_cls: 2.0664 - rpn_regr: 0.3282 - final_cls: 0.9230 - final_regr: 0.4094Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 293/1000 [=======>......................] - ETA: 10:20 - rpn_cls: 2.0659 - rpn_regr: 0.3275 - final_cls: 0.9234 - final_regr: 0.4083Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 301/1000 [========>.....................] - ETA: 10:13 - rpn_cls: 2.0648 - rpn_regr: 0.3271 - final_cls: 0.9236 - final_regr: 0.4076Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 308/1000 [========>.....................] - ETA: 10:06 - rpn_cls: 2.0634 - rpn_regr: 0.3267 - final_cls: 0.9239 - final_regr: 0.4069Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 328/1000 [========>.....................] - ETA: 9:48 - rpn_cls: 2.0589 - rpn_regr: 0.3258 - final_cls: 0.9244 - final_regr: 0.4052Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 329/1000 [========>.....................] - ETA: 9:49 - rpn_cls: 2.0587 - rpn_regr: 0.3257 - final_cls: 0.9245 - final_regr: 0.4051Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 341/1000 [=========>....................] - ETA: 9:39 - rpn_cls: 2.0563 - rpn_regr: 0.3254 - final_cls: 0.9247 - final_regr: 0.4042Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 358/1000 [=========>....................] - ETA: 9:22 - rpn_cls: 2.0546 - rpn_regr: 0.3250 - final_cls: 0.9250 - final_regr: 0.4030Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 376/1000 [==========>...................] - ETA: 9:06 - rpn_cls: 2.0535 - rpn_regr: 0.3248 - final_cls: 0.9253 - final_regr: 0.4019Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 379/1000 [==========>...................] - ETA: 9:05 - rpn_cls: 2.0535 - rpn_regr: 0.3248 - final_cls: 0.9254 - final_regr: 0.4017Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 381/1000 [==========>...................] - ETA: 9:06 - rpn_cls: 2.0535 - rpn_regr: 0.3248 - final_cls: 0.9254 - final_regr: 0.4016Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 383/1000 [==========>...................] - ETA: 9:05 - rpn_cls: 2.0535 - rpn_regr: 0.3248 - final_cls: 0.9255 - final_regr: 0.4015Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 396/1000 [==========>...................] - ETA: 8:53 - rpn_cls: 2.0533 - rpn_regr: 0.3247 - final_cls: 0.9257 - final_regr: 0.4008Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 411/1000 [===========>..................] - ETA: 8:39 - rpn_cls: 2.0537 - rpn_regr: 0.3245 - final_cls: 0.9261 - final_regr: 0.4001Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 436/1000 [============>.................] - ETA: 8:14 - rpn_cls: 2.0554 - rpn_regr: 0.3242 - final_cls: 0.9268 - final_regr: 0.3991Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 450/1000 [============>.................] - ETA: 8:00 - rpn_cls: 2.0566 - rpn_regr: 0.3239 - final_cls: 0.9271 - final_regr: 0.3985Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 483/1000 [=============>................] - ETA: 7:28 - rpn_cls: 2.0585 - rpn_regr: 0.3232 - final_cls: 0.9277 - final_regr: 0.3974Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 504/1000 [==============>...............] - ETA: 7:09 - rpn_cls: 2.0606 - rpn_regr: 0.3227 - final_cls: 0.9279 - final_regr: 0.3967Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 510/1000 [==============>...............] - ETA: 7:04 - rpn_cls: 2.0610 - rpn_regr: 0.3226 - final_cls: 0.9279 - final_regr: 0.3965Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 512/1000 [==============>...............] - ETA: 7:03 - rpn_cls: 2.0612 - rpn_regr: 0.3225 - final_cls: 0.9279 - final_regr: 0.3965Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 528/1000 [==============>...............] - ETA: 6:49 - rpn_cls: 2.0619 - rpn_regr: 0.3223 - final_cls: 0.9280 - final_regr: 0.3960Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 540/1000 [===============>..............] - ETA: 6:38 - rpn_cls: 2.0625 - rpn_regr: 0.3222 - final_cls: 0.9281 - final_regr: 0.3957Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 576/1000 [================>.............] - ETA: 6:04 - rpn_cls: 2.0632 - rpn_regr: 0.3217 - final_cls: 0.9282 - final_regr: 0.3948Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 577/1000 [================>.............] - ETA: 6:03 - rpn_cls: 2.0632 - rpn_regr: 0.3217 - final_cls: 0.9282 - final_regr: 0.3948Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 586/1000 [================>.............] - ETA: 5:56 - rpn_cls: 2.0630 - rpn_regr: 0.3215 - final_cls: 0.9283 - final_regr: 0.3946Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 597/1000 [================>.............] - ETA: 5:46 - rpn_cls: 2.0625 - rpn_regr: 0.3214 - final_cls: 0.9283 - final_regr: 0.3943Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 604/1000 [=================>............] - ETA: 5:40 - rpn_cls: 2.0621 - rpn_regr: 0.3213 - final_cls: 0.9284 - final_regr: 0.3942Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 653/1000 [==================>...........] - ETA: 4:56 - rpn_cls: 2.0600 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3933Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 656/1000 [==================>...........] - ETA: 4:53 - rpn_cls: 2.0601 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3932Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 672/1000 [===================>..........] - ETA: 4:41 - rpn_cls: 2.0604 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3930Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 673/1000 [===================>..........] - ETA: 4:40 - rpn_cls: 2.0604 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3930Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 676/1000 [===================>..........] - ETA: 4:38 - rpn_cls: 2.0605 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3929Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 678/1000 [===================>..........] - ETA: 4:36 - rpn_cls: 2.0606 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3929Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 694/1000 [===================>..........] - ETA: 4:22 - rpn_cls: 2.0610 - rpn_regr: 0.3211 - final_cls: 0.9287 - final_regr: 0.3927Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 700/1000 [====================>.........] - ETA: 4:18 - rpn_cls: 2.0611 - rpn_regr: 0.3212 - final_cls: 0.9288 - final_regr: 0.3927Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 709/1000 [====================>.........] - ETA: 4:10 - rpn_cls: 2.0613 - rpn_regr: 0.3212 - final_cls: 0.9288 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 713/1000 [====================>.........] - ETA: 4:06 - rpn_cls: 2.0614 - rpn_regr: 0.3212 - final_cls: 0.9288 - final_regr: 0.3925Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 782/1000 [======================>.......] - ETA: 3:06 - rpn_cls: 2.0624 - rpn_regr: 0.3211 - final_cls: 0.9288 - final_regr: 0.3918Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 820/1000 [=======================>......] - ETA: 2:33 - rpn_cls: 2.0636 - rpn_regr: 0.3209 - final_cls: 0.9286 - final_regr: 0.3914Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 843/1000 [========================>.....] - ETA: 2:13 - rpn_cls: 2.0647 - rpn_regr: 0.3208 - final_cls: 0.9285 - final_regr: 0.3912Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 874/1000 [=========================>....] - ETA: 1:46 - rpn_cls: 2.0660 - rpn_regr: 0.3207 - final_cls: 0.9283 - final_regr: 0.3909Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 875/1000 [=========================>....] - ETA: 1:46 - rpn_cls: 2.0661 - rpn_regr: 0.3207 - final_cls: 0.9283 - final_regr: 0.3909Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 876/1000 [=========================>....] - ETA: 1:45 - rpn_cls: 2.0661 - rpn_regr: 0.3207 - final_cls: 0.9283 - final_regr: 0.3909Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 887/1000 [=========================>....] - ETA: 1:36 - rpn_cls: 2.0665 - rpn_regr: 0.3206 - final_cls: 0.9282 - final_regr: 0.3908Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 890/1000 [=========================>....] - ETA: 1:33 - rpn_cls: 2.0666 - rpn_regr: 0.3206 - final_cls: 0.9282 - final_regr: 0.3907Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 895/1000 [=========================>....] - ETA: 1:29 - rpn_cls: 2.0668 - rpn_regr: 0.3206 - final_cls: 0.9282 - final_regr: 0.3907Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 922/1000 [==========================>...] - ETA: 1:06 - rpn_cls: 2.0681 - rpn_regr: 0.3204 - final_cls: 0.9280 - final_regr: 0.3905Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 923/1000 [==========================>...] - ETA: 1:05 - rpn_cls: 2.0682 - rpn_regr: 0.3204 - final_cls: 0.9280 - final_regr: 0.3904Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 929/1000 [==========================>...] - ETA: 1:00 - rpn_cls: 2.0685 - rpn_regr: 0.3204 - final_cls: 0.9279 - final_regr: 0.3904Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 948/1000 [===========================>..] - ETA: 44s - rpn_cls: 2.0697 - rpn_regr: 0.3204 - final_cls: 0.9278 - final_regr: 0.3903Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 951/1000 [===========================>..] - ETA: 41s - rpn_cls: 2.0698 - rpn_regr: 0.3204 - final_cls: 0.9278 - final_regr: 0.3902Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 967/1000 [============================>.] - ETA: 28s - rpn_cls: 2.0708 - rpn_regr: 0.3204 - final_cls: 0.9277 - final_regr: 0.3901Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 968/1000 [============================>.] - ETA: 27s - rpn_cls: 2.0709 - rpn_regr: 0.3204 - final_cls: 0.9277 - final_regr: 0.3901Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 988/1000 [============================>.] - ETA: 10s - rpn_cls: 2.0720 - rpn_regr: 0.3204 - final_cls: 0.9276 - final_regr: 0.3900Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 856s 855ms/step - rpn_cls: 2.0726 - rpn_regr: 0.3204 - final_cls: 0.9276 - final_regr: 0.3899\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.807017543859649\n",
            "Classifier accuracy for bounding boxes from RPN: 0.55925\n",
            "Loss RPN classifier: 2.1249199967591337\n",
            "Loss RPN regression: 0.3207303103748709\n",
            "Loss Detector classifier: 0.9244784668385982\n",
            "Loss Detector regression: 0.3828332074023783\n",
            "Total loss: 3.752961981374981\n",
            "Elapsed time: 859.2082850933075\n",
            "Total loss decreased from 3.8309093232081124 to 3.752961981374981, saving weights\n",
            "Epoch 14/40\n",
            "  11/1000 [..............................] - ETA: 13:20 - rpn_cls: 3.2307 - rpn_regr: 0.3020 - final_cls: 1.0139 - final_regr: 0.3789Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  22/1000 [..............................] - ETA: 13:47 - rpn_cls: 2.4574 - rpn_regr: 0.2949 - final_cls: 0.9904 - final_regr: 0.3741Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  30/1000 [..............................] - ETA: 14:18 - rpn_cls: 2.2327 - rpn_regr: 0.2840 - final_cls: 0.9855 - final_regr: 0.3725Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  44/1000 [>.............................] - ETA: 14:11 - rpn_cls: 2.0502 - rpn_regr: 0.2699 - final_cls: 0.9680 - final_regr: 0.3787Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  77/1000 [=>............................] - ETA: 13:15 - rpn_cls: 1.8721 - rpn_regr: 0.2562 - final_cls: 0.9416 - final_regr: 0.3811Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 124/1000 [==>...........................] - ETA: 12:42 - rpn_cls: 1.8143 - rpn_regr: 0.2597 - final_cls: 0.9275 - final_regr: 0.3803Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 136/1000 [===>..........................] - ETA: 12:33 - rpn_cls: 1.8121 - rpn_regr: 0.2617 - final_cls: 0.9266 - final_regr: 0.3798Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 140/1000 [===>..........................] - ETA: 12:32 - rpn_cls: 1.8117 - rpn_regr: 0.2623 - final_cls: 0.9264 - final_regr: 0.3798Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 146/1000 [===>..........................] - ETA: 12:27 - rpn_cls: 1.8114 - rpn_regr: 0.2631 - final_cls: 0.9262 - final_regr: 0.3797Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 168/1000 [====>.........................] - ETA: 12:01 - rpn_cls: 1.8040 - rpn_regr: 0.2663 - final_cls: 0.9246 - final_regr: 0.3790Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 169/1000 [====>.........................] - ETA: 12:03 - rpn_cls: 1.8038 - rpn_regr: 0.2664 - final_cls: 0.9245 - final_regr: 0.3789Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 170/1000 [====>.........................] - ETA: 12:06 - rpn_cls: 1.8036 - rpn_regr: 0.2666 - final_cls: 0.9245 - final_regr: 0.3789Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 179/1000 [====>.........................] - ETA: 12:00 - rpn_cls: 1.8031 - rpn_regr: 0.2677 - final_cls: 0.9241 - final_regr: 0.3786Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 185/1000 [====>.........................] - ETA: 11:54 - rpn_cls: 1.8040 - rpn_regr: 0.2684 - final_cls: 0.9239 - final_regr: 0.3784Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 190/1000 [====>.........................] - ETA: 11:50 - rpn_cls: 1.8043 - rpn_regr: 0.2690 - final_cls: 0.9236 - final_regr: 0.3781Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 196/1000 [====>.........................] - ETA: 11:43 - rpn_cls: 1.8034 - rpn_regr: 0.2696 - final_cls: 0.9232 - final_regr: 0.3779Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 200/1000 [=====>........................] - ETA: 11:41 - rpn_cls: 1.8026 - rpn_regr: 0.2699 - final_cls: 0.9230 - final_regr: 0.3778Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 244/1000 [======>.......................] - ETA: 10:56 - rpn_cls: 1.8036 - rpn_regr: 0.2753 - final_cls: 0.9214 - final_regr: 0.3764Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 247/1000 [======>.......................] - ETA: 10:55 - rpn_cls: 1.8042 - rpn_regr: 0.2757 - final_cls: 0.9214 - final_regr: 0.3763Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 263/1000 [======>.......................] - ETA: 10:44 - rpn_cls: 1.8112 - rpn_regr: 0.2781 - final_cls: 0.9213 - final_regr: 0.3758Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 264/1000 [======>.......................] - ETA: 10:45 - rpn_cls: 1.8119 - rpn_regr: 0.2783 - final_cls: 0.9213 - final_regr: 0.3758Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 287/1000 [=======>......................] - ETA: 10:22 - rpn_cls: 1.8254 - rpn_regr: 0.2817 - final_cls: 0.9218 - final_regr: 0.3753Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 290/1000 [=======>......................] - ETA: 10:23 - rpn_cls: 1.8270 - rpn_regr: 0.2821 - final_cls: 0.9219 - final_regr: 0.3753Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 305/1000 [========>.....................] - ETA: 10:07 - rpn_cls: 1.8360 - rpn_regr: 0.2839 - final_cls: 0.9222 - final_regr: 0.3752Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 374/1000 [==========>...................] - ETA: 8:53 - rpn_cls: 1.8661 - rpn_regr: 0.2895 - final_cls: 0.9225 - final_regr: 0.3749Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 400/1000 [===========>..................] - ETA: 8:33 - rpn_cls: 1.8761 - rpn_regr: 0.2907 - final_cls: 0.9225 - final_regr: 0.3748Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 409/1000 [===========>..................] - ETA: 8:25 - rpn_cls: 1.8800 - rpn_regr: 0.2912 - final_cls: 0.9225 - final_regr: 0.3747Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 418/1000 [===========>..................] - ETA: 8:18 - rpn_cls: 1.8836 - rpn_regr: 0.2917 - final_cls: 0.9224 - final_regr: 0.3746Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 433/1000 [===========>..................] - ETA: 8:05 - rpn_cls: 1.8899 - rpn_regr: 0.2923 - final_cls: 0.9224 - final_regr: 0.3745Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 434/1000 [============>.................] - ETA: 8:05 - rpn_cls: 1.8903 - rpn_regr: 0.2924 - final_cls: 0.9224 - final_regr: 0.3745Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 465/1000 [============>.................] - ETA: 7:35 - rpn_cls: 1.8997 - rpn_regr: 0.2935 - final_cls: 0.9223 - final_regr: 0.3741Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 467/1000 [=============>................] - ETA: 7:34 - rpn_cls: 1.9002 - rpn_regr: 0.2936 - final_cls: 0.9223 - final_regr: 0.3741Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 475/1000 [=============>................] - ETA: 7:28 - rpn_cls: 1.9027 - rpn_regr: 0.2939 - final_cls: 0.9223 - final_regr: 0.3740Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 477/1000 [=============>................] - ETA: 7:27 - rpn_cls: 1.9033 - rpn_regr: 0.2939 - final_cls: 0.9223 - final_regr: 0.3740Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 485/1000 [=============>................] - ETA: 7:20 - rpn_cls: 1.9055 - rpn_regr: 0.2942 - final_cls: 0.9223 - final_regr: 0.3739Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 492/1000 [=============>................] - ETA: 7:14 - rpn_cls: 1.9075 - rpn_regr: 0.2944 - final_cls: 0.9223 - final_regr: 0.3739Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 514/1000 [==============>...............] - ETA: 6:55 - rpn_cls: 1.9133 - rpn_regr: 0.2951 - final_cls: 0.9224 - final_regr: 0.3737Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 540/1000 [===============>..............] - ETA: 6:32 - rpn_cls: 1.9209 - rpn_regr: 0.2962 - final_cls: 0.9223 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 543/1000 [===============>..............] - ETA: 6:29 - rpn_cls: 1.9217 - rpn_regr: 0.2963 - final_cls: 0.9223 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 561/1000 [===============>..............] - ETA: 6:14 - rpn_cls: 1.9268 - rpn_regr: 0.2971 - final_cls: 0.9222 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 564/1000 [===============>..............] - ETA: 6:12 - rpn_cls: 1.9277 - rpn_regr: 0.2972 - final_cls: 0.9222 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 566/1000 [===============>..............] - ETA: 6:11 - rpn_cls: 1.9282 - rpn_regr: 0.2973 - final_cls: 0.9222 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 578/1000 [================>.............] - ETA: 6:01 - rpn_cls: 1.9312 - rpn_regr: 0.2979 - final_cls: 0.9223 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 597/1000 [================>.............] - ETA: 5:44 - rpn_cls: 1.9353 - rpn_regr: 0.2986 - final_cls: 0.9224 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 601/1000 [=================>............] - ETA: 5:41 - rpn_cls: 1.9361 - rpn_regr: 0.2987 - final_cls: 0.9224 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 606/1000 [=================>............] - ETA: 5:37 - rpn_cls: 1.9371 - rpn_regr: 0.2989 - final_cls: 0.9224 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 622/1000 [=================>............] - ETA: 5:23 - rpn_cls: 1.9399 - rpn_regr: 0.2995 - final_cls: 0.9225 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 623/1000 [=================>............] - ETA: 5:22 - rpn_cls: 1.9400 - rpn_regr: 0.2995 - final_cls: 0.9225 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 638/1000 [==================>...........] - ETA: 5:09 - rpn_cls: 1.9420 - rpn_regr: 0.2999 - final_cls: 0.9225 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 696/1000 [===================>..........] - ETA: 4:18 - rpn_cls: 1.9480 - rpn_regr: 0.3012 - final_cls: 0.9223 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 702/1000 [====================>.........] - ETA: 4:13 - rpn_cls: 1.9485 - rpn_regr: 0.3014 - final_cls: 0.9223 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 711/1000 [====================>.........] - ETA: 4:06 - rpn_cls: 1.9492 - rpn_regr: 0.3016 - final_cls: 0.9223 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 723/1000 [====================>.........] - ETA: 3:56 - rpn_cls: 1.9502 - rpn_regr: 0.3019 - final_cls: 0.9222 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 734/1000 [=====================>........] - ETA: 3:46 - rpn_cls: 1.9512 - rpn_regr: 0.3021 - final_cls: 0.9222 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 756/1000 [=====================>........] - ETA: 3:27 - rpn_cls: 1.9528 - rpn_regr: 0.3026 - final_cls: 0.9221 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 759/1000 [=====================>........] - ETA: 3:25 - rpn_cls: 1.9530 - rpn_regr: 0.3027 - final_cls: 0.9221 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 780/1000 [======================>.......] - ETA: 3:07 - rpn_cls: 1.9540 - rpn_regr: 0.3031 - final_cls: 0.9220 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 792/1000 [======================>.......] - ETA: 2:56 - rpn_cls: 1.9545 - rpn_regr: 0.3033 - final_cls: 0.9220 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 807/1000 [=======================>......] - ETA: 2:43 - rpn_cls: 1.9550 - rpn_regr: 0.3036 - final_cls: 0.9220 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 808/1000 [=======================>......] - ETA: 2:43 - rpn_cls: 1.9550 - rpn_regr: 0.3036 - final_cls: 0.9220 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 826/1000 [=======================>......] - ETA: 2:28 - rpn_cls: 1.9558 - rpn_regr: 0.3040 - final_cls: 0.9219 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 837/1000 [========================>.....] - ETA: 2:18 - rpn_cls: 1.9564 - rpn_regr: 0.3042 - final_cls: 0.9219 - final_regr: 0.3733Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 853/1000 [========================>.....] - ETA: 2:05 - rpn_cls: 1.9572 - rpn_regr: 0.3046 - final_cls: 0.9219 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 854/1000 [========================>.....] - ETA: 2:04 - rpn_cls: 1.9573 - rpn_regr: 0.3046 - final_cls: 0.9219 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 858/1000 [========================>.....] - ETA: 2:01 - rpn_cls: 1.9576 - rpn_regr: 0.3047 - final_cls: 0.9219 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 859/1000 [========================>.....] - ETA: 2:00 - rpn_cls: 1.9576 - rpn_regr: 0.3047 - final_cls: 0.9219 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 861/1000 [========================>.....] - ETA: 1:59 - rpn_cls: 1.9578 - rpn_regr: 0.3048 - final_cls: 0.9219 - final_regr: 0.3734Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 883/1000 [=========================>....] - ETA: 1:40 - rpn_cls: 1.9592 - rpn_regr: 0.3052 - final_cls: 0.9219 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 889/1000 [=========================>....] - ETA: 1:35 - rpn_cls: 1.9595 - rpn_regr: 0.3054 - final_cls: 0.9218 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 893/1000 [=========================>....] - ETA: 1:31 - rpn_cls: 1.9596 - rpn_regr: 0.3054 - final_cls: 0.9218 - final_regr: 0.3735Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 902/1000 [==========================>...] - ETA: 1:24 - rpn_cls: 1.9600 - rpn_regr: 0.3056 - final_cls: 0.9218 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 912/1000 [==========================>...] - ETA: 1:15 - rpn_cls: 1.9603 - rpn_regr: 0.3058 - final_cls: 0.9218 - final_regr: 0.3736Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 926/1000 [==========================>...] - ETA: 1:03 - rpn_cls: 1.9607 - rpn_regr: 0.3060 - final_cls: 0.9217 - final_regr: 0.3737Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 930/1000 [==========================>...] - ETA: 1:00 - rpn_cls: 1.9608 - rpn_regr: 0.3061 - final_cls: 0.9217 - final_regr: 0.3737Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 933/1000 [==========================>...] - ETA: 57s - rpn_cls: 1.9609 - rpn_regr: 0.3061 - final_cls: 0.9217 - final_regr: 0.3737Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 971/1000 [============================>.] - ETA: 24s - rpn_cls: 1.9617 - rpn_regr: 0.3065 - final_cls: 0.9215 - final_regr: 0.3738Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 977/1000 [============================>.] - ETA: 19s - rpn_cls: 1.9618 - rpn_regr: 0.3066 - final_cls: 0.9215 - final_regr: 0.3738Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 982/1000 [============================>.] - ETA: 15s - rpn_cls: 1.9619 - rpn_regr: 0.3067 - final_cls: 0.9215 - final_regr: 0.3738Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 857s 857ms/step - rpn_cls: 1.9620 - rpn_regr: 0.3069 - final_cls: 0.9214 - final_regr: 0.3739\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 8.973394495412844\n",
            "Classifier accuracy for bounding boxes from RPN: 0.5655\n",
            "Loss RPN classifier: 1.9697696193749539\n",
            "Loss RPN regression: 0.3185139198433608\n",
            "Loss Detector classifier: 0.917189157128334\n",
            "Loss Detector regression: 0.37749312646035105\n",
            "Total loss: 3.5829658228069996\n",
            "Elapsed time: 859.7221598625183\n",
            "Total loss decreased from 3.752961981374981 to 3.5829658228069996, saving weights\n",
            "Epoch 15/40\n",
            "  14/1000 [..............................] - ETA: 13:36 - rpn_cls: 2.3020 - rpn_regr: 0.3503 - final_cls: 1.0666 - final_regr: 0.4370Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  25/1000 [..............................] - ETA: 13:41 - rpn_cls: 2.2269 - rpn_regr: 0.3191 - final_cls: 1.0297 - final_regr: 0.4087Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  47/1000 [>.............................] - ETA: 13:19 - rpn_cls: 2.0863 - rpn_regr: 0.3108 - final_cls: 1.0000 - final_regr: 0.4005Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  48/1000 [>.............................] - ETA: 13:38 - rpn_cls: 2.0797 - rpn_regr: 0.3105 - final_cls: 0.9995 - final_regr: 0.4003Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  49/1000 [>.............................] - ETA: 13:53 - rpn_cls: 2.0732 - rpn_regr: 0.3102 - final_cls: 0.9990 - final_regr: 0.4002Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  65/1000 [>.............................] - ETA: 13:41 - rpn_cls: 1.9549 - rpn_regr: 0.3063 - final_cls: 0.9970 - final_regr: 0.4043Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  77/1000 [=>............................] - ETA: 13:25 - rpn_cls: 1.8829 - rpn_regr: 0.3038 - final_cls: 0.9931 - final_regr: 0.4069Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  97/1000 [=>............................] - ETA: 13:06 - rpn_cls: 1.8172 - rpn_regr: 0.3020 - final_cls: 0.9841 - final_regr: 0.4079Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 125/1000 [==>...........................] - ETA: 12:49 - rpn_cls: 1.7917 - rpn_regr: 0.3071 - final_cls: 0.9780 - final_regr: 0.4089Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 128/1000 [==>...........................] - ETA: 12:49 - rpn_cls: 1.7923 - rpn_regr: 0.3077 - final_cls: 0.9774 - final_regr: 0.4089Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 144/1000 [===>..........................] - ETA: 12:41 - rpn_cls: 1.8042 - rpn_regr: 0.3106 - final_cls: 0.9745 - final_regr: 0.4087Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 145/1000 [===>..........................] - ETA: 12:45 - rpn_cls: 1.8052 - rpn_regr: 0.3108 - final_cls: 0.9744 - final_regr: 0.4087Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 154/1000 [===>..........................] - ETA: 12:39 - rpn_cls: 1.8158 - rpn_regr: 0.3125 - final_cls: 0.9732 - final_regr: 0.4088Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 183/1000 [====>.........................] - ETA: 12:11 - rpn_cls: 1.8423 - rpn_regr: 0.3161 - final_cls: 0.9703 - final_regr: 0.4095Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 187/1000 [====>.........................] - ETA: 12:08 - rpn_cls: 1.8467 - rpn_regr: 0.3164 - final_cls: 0.9700 - final_regr: 0.4095Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 192/1000 [====>.........................] - ETA: 12:05 - rpn_cls: 1.8528 - rpn_regr: 0.3167 - final_cls: 0.9697 - final_regr: 0.4094Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 195/1000 [====>.........................] - ETA: 12:04 - rpn_cls: 1.8561 - rpn_regr: 0.3169 - final_cls: 0.9695 - final_regr: 0.4094Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 237/1000 [======>.......................] - ETA: 11:13 - rpn_cls: 1.8874 - rpn_regr: 0.3174 - final_cls: 0.9661 - final_regr: 0.4084Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 253/1000 [======>.......................] - ETA: 10:56 - rpn_cls: 1.8943 - rpn_regr: 0.3170 - final_cls: 0.9644 - final_regr: 0.4077Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 279/1000 [=======>......................] - ETA: 10:33 - rpn_cls: 1.9033 - rpn_regr: 0.3162 - final_cls: 0.9616 - final_regr: 0.4063Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 314/1000 [========>.....................] - ETA: 9:57 - rpn_cls: 1.9116 - rpn_regr: 0.3154 - final_cls: 0.9587 - final_regr: 0.4041Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 322/1000 [========>.....................] - ETA: 9:49 - rpn_cls: 1.9125 - rpn_regr: 0.3152 - final_cls: 0.9580 - final_regr: 0.4037Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 344/1000 [=========>....................] - ETA: 9:26 - rpn_cls: 1.9137 - rpn_regr: 0.3147 - final_cls: 0.9563 - final_regr: 0.4027Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 345/1000 [=========>....................] - ETA: 9:26 - rpn_cls: 1.9137 - rpn_regr: 0.3147 - final_cls: 0.9562 - final_regr: 0.4027Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 346/1000 [=========>....................] - ETA: 9:27 - rpn_cls: 1.9137 - rpn_regr: 0.3146 - final_cls: 0.9561 - final_regr: 0.4026Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 354/1000 [=========>....................] - ETA: 9:23 - rpn_cls: 1.9134 - rpn_regr: 0.3144 - final_cls: 0.9555 - final_regr: 0.4022Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 373/1000 [==========>...................] - ETA: 9:05 - rpn_cls: 1.9113 - rpn_regr: 0.3141 - final_cls: 0.9542 - final_regr: 0.4013Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 393/1000 [==========>...................] - ETA: 8:47 - rpn_cls: 1.9087 - rpn_regr: 0.3138 - final_cls: 0.9528 - final_regr: 0.4004Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 394/1000 [==========>...................] - ETA: 8:47 - rpn_cls: 1.9086 - rpn_regr: 0.3137 - final_cls: 0.9528 - final_regr: 0.4004Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 420/1000 [===========>..................] - ETA: 8:23 - rpn_cls: 1.9087 - rpn_regr: 0.3137 - final_cls: 0.9512 - final_regr: 0.3995Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 423/1000 [===========>..................] - ETA: 8:21 - rpn_cls: 1.9088 - rpn_regr: 0.3137 - final_cls: 0.9510 - final_regr: 0.3994Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 430/1000 [===========>..................] - ETA: 8:17 - rpn_cls: 1.9090 - rpn_regr: 0.3138 - final_cls: 0.9506 - final_regr: 0.3991Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 440/1000 [============>.................] - ETA: 8:09 - rpn_cls: 1.9096 - rpn_regr: 0.3138 - final_cls: 0.9500 - final_regr: 0.3988Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 446/1000 [============>.................] - ETA: 8:05 - rpn_cls: 1.9102 - rpn_regr: 0.3139 - final_cls: 0.9497 - final_regr: 0.3986Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 463/1000 [============>.................] - ETA: 7:50 - rpn_cls: 1.9126 - rpn_regr: 0.3141 - final_cls: 0.9489 - final_regr: 0.3982Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 471/1000 [=============>................] - ETA: 7:43 - rpn_cls: 1.9137 - rpn_regr: 0.3142 - final_cls: 0.9486 - final_regr: 0.3980Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 477/1000 [=============>................] - ETA: 7:38 - rpn_cls: 1.9146 - rpn_regr: 0.3142 - final_cls: 0.9484 - final_regr: 0.3978Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 487/1000 [=============>................] - ETA: 7:28 - rpn_cls: 1.9158 - rpn_regr: 0.3143 - final_cls: 0.9480 - final_regr: 0.3976Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 490/1000 [=============>................] - ETA: 7:26 - rpn_cls: 1.9162 - rpn_regr: 0.3143 - final_cls: 0.9479 - final_regr: 0.3975Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 491/1000 [=============>................] - ETA: 7:26 - rpn_cls: 1.9163 - rpn_regr: 0.3143 - final_cls: 0.9478 - final_regr: 0.3975Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 503/1000 [==============>...............] - ETA: 7:16 - rpn_cls: 1.9175 - rpn_regr: 0.3143 - final_cls: 0.9474 - final_regr: 0.3971Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 513/1000 [==============>...............] - ETA: 7:07 - rpn_cls: 1.9182 - rpn_regr: 0.3143 - final_cls: 0.9470 - final_regr: 0.3969Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 529/1000 [==============>...............] - ETA: 6:51 - rpn_cls: 1.9189 - rpn_regr: 0.3141 - final_cls: 0.9464 - final_regr: 0.3965Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 545/1000 [===============>..............] - ETA: 6:36 - rpn_cls: 1.9192 - rpn_regr: 0.3140 - final_cls: 0.9457 - final_regr: 0.3962Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 575/1000 [================>.............] - ETA: 6:09 - rpn_cls: 1.9198 - rpn_regr: 0.3136 - final_cls: 0.9444 - final_regr: 0.3954Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 581/1000 [================>.............] - ETA: 6:05 - rpn_cls: 1.9200 - rpn_regr: 0.3136 - final_cls: 0.9441 - final_regr: 0.3952Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 608/1000 [=================>............] - ETA: 5:40 - rpn_cls: 1.9211 - rpn_regr: 0.3133 - final_cls: 0.9430 - final_regr: 0.3946Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 639/1000 [==================>...........] - ETA: 5:11 - rpn_cls: 1.9209 - rpn_regr: 0.3130 - final_cls: 0.9416 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 640/1000 [==================>...........] - ETA: 5:11 - rpn_cls: 1.9208 - rpn_regr: 0.3130 - final_cls: 0.9416 - final_regr: 0.3940Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 649/1000 [==================>...........] - ETA: 5:03 - rpn_cls: 1.9205 - rpn_regr: 0.3130 - final_cls: 0.9412 - final_regr: 0.3938Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 660/1000 [==================>...........] - ETA: 4:54 - rpn_cls: 1.9199 - rpn_regr: 0.3129 - final_cls: 0.9407 - final_regr: 0.3937Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 672/1000 [===================>..........] - ETA: 4:43 - rpn_cls: 1.9192 - rpn_regr: 0.3128 - final_cls: 0.9403 - final_regr: 0.3935Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 689/1000 [===================>..........] - ETA: 4:29 - rpn_cls: 1.9184 - rpn_regr: 0.3127 - final_cls: 0.9396 - final_regr: 0.3932Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 703/1000 [====================>.........] - ETA: 4:16 - rpn_cls: 1.9182 - rpn_regr: 0.3126 - final_cls: 0.9391 - final_regr: 0.3930Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 714/1000 [====================>.........] - ETA: 4:07 - rpn_cls: 1.9182 - rpn_regr: 0.3126 - final_cls: 0.9387 - final_regr: 0.3928Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 717/1000 [====================>.........] - ETA: 4:05 - rpn_cls: 1.9182 - rpn_regr: 0.3126 - final_cls: 0.9386 - final_regr: 0.3928Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 733/1000 [====================>.........] - ETA: 3:51 - rpn_cls: 1.9185 - rpn_regr: 0.3127 - final_cls: 0.9381 - final_regr: 0.3926Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 758/1000 [=====================>........] - ETA: 3:30 - rpn_cls: 1.9194 - rpn_regr: 0.3128 - final_cls: 0.9374 - final_regr: 0.3923Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 773/1000 [======================>.......] - ETA: 3:17 - rpn_cls: 1.9201 - rpn_regr: 0.3129 - final_cls: 0.9370 - final_regr: 0.3922Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 777/1000 [======================>.......] - ETA: 3:13 - rpn_cls: 1.9202 - rpn_regr: 0.3129 - final_cls: 0.9369 - final_regr: 0.3921Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 782/1000 [======================>.......] - ETA: 3:09 - rpn_cls: 1.9205 - rpn_regr: 0.3129 - final_cls: 0.9367 - final_regr: 0.3921Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 785/1000 [======================>.......] - ETA: 3:06 - rpn_cls: 1.9206 - rpn_regr: 0.3129 - final_cls: 0.9367 - final_regr: 0.3921Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 810/1000 [=======================>......] - ETA: 2:44 - rpn_cls: 1.9213 - rpn_regr: 0.3130 - final_cls: 0.9361 - final_regr: 0.3918Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 857/1000 [========================>.....] - ETA: 2:03 - rpn_cls: 1.9221 - rpn_regr: 0.3129 - final_cls: 0.9348 - final_regr: 0.3913Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 863/1000 [========================>.....] - ETA: 1:58 - rpn_cls: 1.9223 - rpn_regr: 0.3128 - final_cls: 0.9347 - final_regr: 0.3912Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 872/1000 [=========================>....] - ETA: 1:50 - rpn_cls: 1.9224 - rpn_regr: 0.3128 - final_cls: 0.9344 - final_regr: 0.3911Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 936/1000 [===========================>..] - ETA: 54s - rpn_cls: 1.9231 - rpn_regr: 0.3127 - final_cls: 0.9329 - final_regr: 0.3903Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 937/1000 [===========================>..] - ETA: 54s - rpn_cls: 1.9231 - rpn_regr: 0.3127 - final_cls: 0.9329 - final_regr: 0.3903Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 938/1000 [===========================>..] - ETA: 53s - rpn_cls: 1.9231 - rpn_regr: 0.3127 - final_cls: 0.9328 - final_regr: 0.3903Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 967/1000 [============================>.] - ETA: 28s - rpn_cls: 1.9233 - rpn_regr: 0.3127 - final_cls: 0.9322 - final_regr: 0.3901Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 987/1000 [============================>.] - ETA: 11s - rpn_cls: 1.9233 - rpn_regr: 0.3126 - final_cls: 0.9318 - final_regr: 0.3899Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 988/1000 [============================>.] - ETA: 10s - rpn_cls: 1.9233 - rpn_regr: 0.3126 - final_cls: 0.9317 - final_regr: 0.3899Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "1000/1000 [==============================] - 860s 859ms/step - rpn_cls: 1.9235 - rpn_regr: 0.3126 - final_cls: 0.9315 - final_regr: 0.3898\n",
            "Mean number of bounding boxes from RPN overlapping ground truth boxes: 9.13531047265987\n",
            "Classifier accuracy for bounding boxes from RPN: 0.57325\n",
            "Loss RPN classifier: 1.946259407128468\n",
            "Loss RPN regression: 0.3130759754450992\n",
            "Loss Detector classifier: 0.9122828031778336\n",
            "Loss Detector regression: 0.3832709804344922\n",
            "Total loss: 3.554889166185893\n",
            "Elapsed time: 862.5658593177795\n",
            "Total loss decreased from 3.5829658228069996 to 3.554889166185893, saving weights\n",
            "Epoch 16/40\n",
            "Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  25/1000 [..............................] - ETA: 13:35 - rpn_cls: 2.1756 - rpn_regr: 0.3982 - final_cls: 0.9422 - final_regr: 0.3081Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  35/1000 [>.............................] - ETA: 13:58 - rpn_cls: 2.1889 - rpn_regr: 0.3995 - final_cls: 0.9367 - final_regr: 0.3132Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  38/1000 [>.............................] - ETA: 14:21 - rpn_cls: 2.1916 - rpn_regr: 0.3988 - final_cls: 0.9370 - final_regr: 0.3163Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  53/1000 [>.............................] - ETA: 14:22 - rpn_cls: 2.1789 - rpn_regr: 0.3942 - final_cls: 0.9435 - final_regr: 0.3293Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  56/1000 [>.............................] - ETA: 14:38 - rpn_cls: 2.1679 - rpn_regr: 0.3925 - final_cls: 0.9442 - final_regr: 0.3313Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  59/1000 [>.............................] - ETA: 14:53 - rpn_cls: 2.1619 - rpn_regr: 0.3905 - final_cls: 0.9442 - final_regr: 0.3333Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            "  70/1000 [=>............................] - ETA: 14:32 - rpn_cls: 2.1596 - rpn_regr: 0.3826 - final_cls: 0.9432 - final_regr: 0.3389Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 144/1000 [===>..........................] - ETA: 12:26 - rpn_cls: 2.1253 - rpn_regr: 0.3462 - final_cls: 0.9393 - final_regr: 0.3485Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 157/1000 [===>..........................] - ETA: 12:20 - rpn_cls: 2.1179 - rpn_regr: 0.3422 - final_cls: 0.9380 - final_regr: 0.3488Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 180/1000 [====>.........................] - ETA: 11:58 - rpn_cls: 2.1130 - rpn_regr: 0.3368 - final_cls: 0.9360 - final_regr: 0.3491Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 189/1000 [====>.........................] - ETA: 11:52 - rpn_cls: 2.1111 - rpn_regr: 0.3352 - final_cls: 0.9350 - final_regr: 0.3492Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 213/1000 [=====>........................] - ETA: 11:21 - rpn_cls: 2.1063 - rpn_regr: 0.3311 - final_cls: 0.9336 - final_regr: 0.3502Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 235/1000 [======>.......................] - ETA: 10:57 - rpn_cls: 2.1029 - rpn_regr: 0.3280 - final_cls: 0.9320 - final_regr: 0.3509Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 236/1000 [======>.......................] - ETA: 10:59 - rpn_cls: 2.1029 - rpn_regr: 0.3279 - final_cls: 0.9319 - final_regr: 0.3509Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 238/1000 [======>.......................] - ETA: 11:00 - rpn_cls: 2.1030 - rpn_regr: 0.3276 - final_cls: 0.9318 - final_regr: 0.3510Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 246/1000 [======>.......................] - ETA: 10:53 - rpn_cls: 2.1033 - rpn_regr: 0.3266 - final_cls: 0.9314 - final_regr: 0.3513Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 265/1000 [======>.......................] - ETA: 10:33 - rpn_cls: 2.1015 - rpn_regr: 0.3245 - final_cls: 0.9309 - final_regr: 0.3521Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 285/1000 [=======>......................] - ETA: 10:15 - rpn_cls: 2.0967 - rpn_regr: 0.3228 - final_cls: 0.9305 - final_regr: 0.3530Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 286/1000 [=======>......................] - ETA: 10:17 - rpn_cls: 2.0965 - rpn_regr: 0.3227 - final_cls: 0.9305 - final_regr: 0.3530Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 324/1000 [========>.....................] - ETA: 9:42 - rpn_cls: 2.0946 - rpn_regr: 0.3211 - final_cls: 0.9300 - final_regr: 0.3549Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 334/1000 [=========>....................] - ETA: 9:33 - rpn_cls: 2.0950 - rpn_regr: 0.3211 - final_cls: 0.9299 - final_regr: 0.3554Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 338/1000 [=========>....................] - ETA: 9:31 - rpn_cls: 2.0953 - rpn_regr: 0.3211 - final_cls: 0.9298 - final_regr: 0.3555Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 342/1000 [=========>....................] - ETA: 9:28 - rpn_cls: 2.0956 - rpn_regr: 0.3211 - final_cls: 0.9296 - final_regr: 0.3557Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 371/1000 [==========>...................] - ETA: 9:02 - rpn_cls: 2.0988 - rpn_regr: 0.3208 - final_cls: 0.9288 - final_regr: 0.3568Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 375/1000 [==========>...................] - ETA: 8:58 - rpn_cls: 2.0990 - rpn_regr: 0.3207 - final_cls: 0.9287 - final_regr: 0.3569Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 380/1000 [==========>...................] - ETA: 8:55 - rpn_cls: 2.0991 - rpn_regr: 0.3206 - final_cls: 0.9287 - final_regr: 0.3571Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 394/1000 [==========>...................] - ETA: 8:43 - rpn_cls: 2.0993 - rpn_regr: 0.3204 - final_cls: 0.9284 - final_regr: 0.3575Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 408/1000 [===========>..................] - ETA: 8:29 - rpn_cls: 2.0989 - rpn_regr: 0.3200 - final_cls: 0.9281 - final_regr: 0.3579Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 409/1000 [===========>..................] - ETA: 8:29 - rpn_cls: 2.0988 - rpn_regr: 0.3200 - final_cls: 0.9281 - final_regr: 0.3579Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 440/1000 [============>.................] - ETA: 7:59 - rpn_cls: 2.0972 - rpn_regr: 0.3189 - final_cls: 0.9274 - final_regr: 0.3588Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 460/1000 [============>.................] - ETA: 7:42 - rpn_cls: 2.0958 - rpn_regr: 0.3183 - final_cls: 0.9269 - final_regr: 0.3591Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 465/1000 [============>.................] - ETA: 7:39 - rpn_cls: 2.0953 - rpn_regr: 0.3181 - final_cls: 0.9268 - final_regr: 0.3592Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 497/1000 [=============>................] - ETA: 7:10 - rpn_cls: 2.0911 - rpn_regr: 0.3173 - final_cls: 0.9260 - final_regr: 0.3596Exception: in user code:\n",
            "\n",
            "    File \"/usr/local/lib/python3.7/dist-packages/keras/engine/training.py\", line 878, in train_function  *\n",
            "        return step_function(self, iterator)\n",
            "    File \"<ipython-input-17-84f9950f7cd7>\", line 53, in class_loss_regr_fixed_num  *\n",
            "        x = y_true[:, :, 4*num_classes:] - y_pred\n",
            "\n",
            "    TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'.\n",
            "\n",
            " 524/1000 [==============>...............] - ETA: 6:45 - rpn_cls: 2.0864 - rpn_regr: 0.3167 - final_cls: 0.9254 - final_regr: 0.3599"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Kt-1Grs90oD3"
      },
      "source": [
        "plt.figure(figsize=(15,5))\n",
        "plt.subplot(1,2,1)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['mean_overlapping_bboxes'], 'r')\n",
        "plt.title('mean_overlapping_bboxes')\n",
        "plt.subplot(1,2,2)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['class_acc'], 'r')\n",
        "plt.title('class_acc')\n",
        "\n",
        "plt.show()\n",
        "\n",
        "plt.figure(figsize=(15,5))\n",
        "plt.subplot(1,2,1)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['loss_rpn_cls'], 'r')\n",
        "plt.title('loss_rpn_cls')\n",
        "plt.subplot(1,2,2)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['loss_rpn_regr'], 'r')\n",
        "plt.title('loss_rpn_regr')\n",
        "plt.show()\n",
        "\n",
        "\n",
        "plt.figure(figsize=(15,5))\n",
        "plt.subplot(1,2,1)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['loss_class_cls'], 'r')\n",
        "plt.title('loss_class_cls')\n",
        "plt.subplot(1,2,2)\n",
        "plt.plot(np.arange(0, r_epochs), record_df['loss_class_regr'], 'r')\n",
        "plt.title('loss_class_regr')\n",
        "plt.show()\n",
        "\n",
        "plt.plot(np.arange(0, r_epochs), record_df['curr_loss'], 'r')\n",
        "plt.title('total_loss')\n",
        "plt.show()\n",
        "\n",
        "# plt.figure(figsize=(15,5))\n",
        "# plt.subplot(1,2,1)\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['curr_loss'], 'r')\n",
        "# plt.title('total_loss')\n",
        "# plt.subplot(1,2,2)\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['elapsed_time'], 'r')\n",
        "# plt.title('elapsed_time')\n",
        "# plt.show()\n",
        "\n",
        "# plt.title('loss')\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['loss_rpn_cls'], 'b')\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['loss_rpn_regr'], 'g')\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['loss_class_cls'], 'r')\n",
        "# plt.plot(np.arange(0, r_epochs), record_df['loss_class_regr'], 'c')\n",
        "# # plt.plot(np.arange(0, r_epochs), record_df['curr_loss'], 'm')\n",
        "# plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-fgI0ovv3zlr"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}