NPU_CaffeSSD

RK3399 Pro NPU support for Caffe SSD detector

assume you have installed RKNN toolkit and it's dependencies or you can install it first by

$ pip install rknn_toolkit-0.9.9-cp35-cp35m-linux_x86_64.whl

the package you can find at

1. [Main Site]
2. [Download Source Site]

you can find the pedestrian detecttion model of this project at [here].

Usage

extract priorbox (you need change the caffe_root path to yours in priorbox.py, or may need install caffe-ssd first, caffe-ssd you can find at [here])

$ python priorbox.py

convert model and run pedestrian detection (may take some time to run)

$ python3 npu_ssd_det.py

Result

you can use NMS algorithm to remove redundant rects.

Issues

1. In order to get priorbox ,when you change the detection model, you should modify model's prototxt looks like MobileNetSSD_deploy_truncated.prototxt

2. Different model has different outlen (npu_ssd_det.py),If you change the detection model, you should replace the outlen first,How to do this see [issue]