rmcong / FPNet_ACMMM23

12 stars 1 forks source link

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

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