Towards Open-World Co-Salient Object Detection with Generative Uncertainty-aware Group Selective Exchange-Masking

Thank you very much for your interest in our work.

The OWCoSOD datasets (OWCoSal, OWCoSOD, OWCoCA) can be downloaded from the link
https://pan.baidu.com/s/11MKqPIRP58p8lvz7x9AF2Q, and the password is 1310.

The results of our method on OWCoSal, OWCoSOD, OWCoCA can be downloaded from
https://pan.baidu.com/s/1Yw3jN_cxkRgR47PSiIclPw, and the password is 1310.

The contour can be generated by draw_contour.py.

Co-Salient Object Detection with Uncertainty-aware Group Exchange-Masking

The results of our method on CoSal2015, CoSOD3k, CoCA, MSRC, and iCoseg are available, and they can be downloaded from the link
https://pan.baidu.com/s/1uRwH5Y1HgDvxWd9gwRoR9g, and the password is 1310.

The pretrained_model and weights can be downloaded from the link
https://pan.baidu.com/s/1_FVoR6QP6FeQCZEGxgyZog and the password is 1310.

Putting the pretrained_model into ./pretrained_model and weights files into ./result/models and run coseg_test.py can get the results.

The link of the eval toolbox is: https://github.com/zzhanghub/eval-co-sod, we are very grateful for their contributions.

Installation Instructions

- Python 3.8
- Pytorch 2.1.1(CUDA 12.1 build).

Please see requirements.txt for all the other requirements.

Train on coco-seg

When you initially train the method, you need firstly train vqvae and pixelcnn

Train vqvae

python train_VQVAE.py

Train pixelcnn

python train_pixelcnn.py

Train our Method

python main.py --data_root /home/dell/Codes/IJCV/data/  --trainset coco-seg --n_embedding 128 --n_dim 384 --color_level 128 --linear_dim 128 --save_vqvae ./checkpoints/vqvae --save_gen_model ./checkpoints/vqvae

Test on CoCA, CoSOD3k, CoSal2015

Test on CoCA

python coseg_test.py --n_embedding 128 --n_dim 384 --color_level 128 --linear_dim 128

PS

If you find our work helpful, you can cite our paper

@inproceedings{wu2023co,
  title={Co-Salient Object Detection With Uncertainty-Aware Group Exchange-Masking},
  author={Wu, Yang and Song, Huihui and Liu, Bo and Zhang, Kaihua and Liu, Dong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={19639--19648},
  year={2023}
}
@article{wu2023towards,
  title={Towards Open-World Co-Salient Object Detection with Generative Uncertainty-aware Group Selective Exchange-Masking},
  author={Wu, Yang and Hu, Shenglong and Song, Huihui and Zhang, Kaihua and Liu, Bo and Liu, Dong},
  journal={arXiv preprint arXiv:2310.10264},
  year={2023}
}