ras-pytorch's Introduction

RAS-pytorch

The pytorch code for our TIP2020 paper: Reverse Attention Based Residual Network for Salient Object Detection

Notice

We use ResNet50 as backbone and add the IoU loss[1] for better performance in this pytorch implementation. The original caffe version is here. We provide two versions with different training strategies.

v1: The same training strategy with CPD[2].
v2: The same training strategy with F3Net[3].

Usage

Modify the pathes of datasets, then run:

Training: python3 train.py
Testing:  python3 test.py

Performace

The codes are tested on Ubuntu 18.04 environment (Python3.6.9, PyTorch1.5.0, torchvision0.6.0, CUDA10.2, cuDNN7.6.5) with RTX 2080Ti GPU. We select several recent SOTA methods for comparisons. The evaluation code can be found here.

Pre-trained models & Pre-computed saliency maps

v1: model Baidu(bc3k) Google; smaps Baidu(kp6t) Google
v2: model Baidu(wbz1) Google; smaps Baidu(j57z) Google

Citation

@article{chen2020tip, 
  author={Chen, Shuhan and Tan, Xiuli and Wang, Ben and Lu, Huchuan and Hu, Xuelong and Fu, Yun}, 
  journal={IEEE Transactions on Image Processing}, 
  title={Reverse Attention-Based Residual Network for Salient Object Detection},
  volume={29},  
  pages={3763-3776},
  year={2020}
}

@inproceedings{chen2018eccv, 
  author={Chen, Shuhan and Tan, Xiuli and Wang, Ben and Hu, Xuelong}, 
  booktitle={European Conference on Computer Vision}, 
  title={Reverse Attention for Salient Object Detection}, 
  year={2018}
}

Acknowledgements

This code is built on CPD[2] and F3Net[3]. We thank the authors for sharing their codes.

Reference

[1] Xuebin Qin, Zichen Zhang, Chenyang Huang, Chao Gao, Masood Dehghan, Martin Jagersand. BASNet: Boundary-Aware Salient Object Detection. In CVPR, 2019.
[2] Zhe Wu, Li Su, Qingming Huang. Cascaded Partial Decoder for Fast and Accurate Salient Object Detection. In CVPR, 2019.
[3] Jun Wei, Shuhui Wang, Qingming Huang. F3Net: Fusion, Feedback and Focus for Salient Object Detection. In AAAI, 2020.

Recommend Projects