Towards End-to-End Unsupervised Saliency Detection with Self-Supervised Top-Down Context

Requirements

• Python 3.6
• Pytorch 1.7
• Install submodule Connected_components_PyTorch
• Prepare the pretrained Moco-v2 weight

Usage

# For training
python train.py    

# For testing a saved model weight from a specific epoch
python test_tool.py  

Acknowledgement

Our idea is inspired by PSOD, A2S-v2, C2AM, and AFA. We also thank Connected_components_PyTorch for providing a high-performance algorithm and implementation for calculating connected components.

Citation

@inproceedings{song2023towards,
  title={Towards End-to-End Unsupervised Saliency Detection with Self-Supervised Top-Down Context},
  author={Song, Yicheng and Gao, Shuyong and Xing, Haozhe and Cheng, Yiting and Wang, Yan and Zhang, Wenqiang},
  booktitle={Proceedings of the 31st ACM International Conference on Multimedia},
  pages={5532--5541},
  year={2023}
}