Frequency Perception Network for Camouflaged Object Detection

Authors: Runmin Cong, Mengyao Sun, Sanyi Zhang, Xiaofei Zhou, Wei Zhang, and Yao Zhao.

The source code can be found from Baidu Drive，CODE: MVPL

1. Preface

• This repository provides code for "Frequency Perception Network for Camouflaged Object Detection" ACM MM 2023. Paper

2. Proposed Method

2.1. Training/Testing

The training and testing experiments are conducted using PyTorch with one NVIDIA 2080Ti GPU of 32 GB Memory.

1. Configuring your environment (Prerequisites):

   • Installing necessary packages:

     python 3.6

     torch 1.11.0

     numpy 1.22.4

     mmcv-full 1.7.1

     timm 0.6.13

     mmdet 2.19.1

2. Downloading necessary data:

   • downloading dataset and move it into ./data/, which can be found from Baidu Drive.

   • downloading our weights and move it into ./snapshot/FPNet-GroupInsert/FPNet.pth.

   • downloading PVTv1-Large weights and move it into ./lib/models/pvt_large.pth.

3. Training Configuration:

   • After you download training dataset, just run MyTrain_Val.py to train our model.

4. Testing Configuration:

   • After you download all the pre-trained model and testing dataset, just run MyTesting.py to generate the final prediction maps.

   • You can also download prediction maps ('CHAMELEON', 'CAMO', 'COD10K') from Baidu Drive.

2.2 Evaluating your trained model:

One evaluation is written in Python code please follow this the instructions in ./evaluator.py and just run it to generate the evaluation results in. We implement four metrics: MAE (Mean Absolute Error), weighted F-measure, mean E-measure, S-Measure.

⬆ back to top