Home

Awesome

Improving Camouflaged Object Detection with the Uncertainty of Pseudo-edge Labels (UR-COD)

This is the official implementation of the paper in ACM Multimedia Asia 2021. We provide the sample codes for training and testing and pretrained models on camouflaged object detection.

<p align="left"> <img src="figure/proposed_method.png" alt="architecture" width="875px"> </p>

Requirements

Installation

  1. Clone the repository.
git clone git@github.com:nobukatsu-kajiura/UR-COD.git
  1. Download dataset from here. Put TrainDataset and TestDataset directory in $ROOT/data/.

  2. If you want to try the framework, download the pretrained model from here and put Model_100_gen.pth in $ROOT/checkpoints/UR-SINetv2-pretrained/.

Training and testing

  1. Train the framework.
python train.py --name UR-SINetv2 --dataset SINetv2 --epoch 100
  1. Test the framework. When using the pretrained model, use --name UR-SINetv2-pretrained.
python test.py --name UR-SINetv2 --dataset SINetv2 --epoch 100

References

We used the publicly avaliable model of SINet-v2[Fan+, TPAMI2021] as Pseudo-Map Generator. We obtained the weights from here and generated pseudo-map labels.

Citation

If you find our research useful in your research, please consider citing:

@inproceedings{kajiura2021improving,
    title={Improving Camouflaged Object Detection with the Uncertainty of Pseudo-edge Labels},
    author={Kajiura, Nobukatsu and Liu, Hong and Satoh, Shin'ichi},
    booktitle = {ACM Multimedia Asia},
    year = {2021}
}