juampamuc / active-boundary-loss

Official repository for Active Boundary Loss for Semantic Segmentation.

active-boundary-loss

---

By Chi Wang, Yunke Zhang, Miaomiao Cui, Peiran Ren, Yin Yang, Xuansong Xie, Xian-Sheng Hua, Hujun Bao and Weiwei Xu.

This repo is the official implementation of Active Boundary Loss for Semantic Segmentation (Paper link; arXiv:2102.02696) accepted by AAAI 2022 (Oral).

Dependencies

• Python libarary:
  • pytorch>=1.0
  • torchvision
  • numpy
  • scipy
  • opencv-python
• label smooth
• [option] Lovász-Softmax loss

How to use

• Directly use ABL in abl.py.
• For ease of use, we provide an all-in-one version. In addition, you can speed it up by moving the process of distance map generation to the dataloader.

Citation

@inproceedings{wang2022active,
  title={Active boundary loss for semantic segmentation},
  author={Wang, Chi and Zhang, Yunke and Cui, Miaomiao and Ren, Peiran and Yang, Yin and Xie, Xuansong and Hua, Xian-Sheng and Bao, Hujun and Xu, Weiwei},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={36},
  number={2},
  pages={2397--2405},
  year={2022}
}

Contact

If you have any questions, please feel free to contact [email protected].