CECL

0.Citation

If you find our work inspiring or use our codebase in your research, please consider giving a star ⭐ and a citation.

@inproceedings{wan2024unlocking,
  title={Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning},
  author={Wan, Wenhai and Wang, Xinrui and Xie, Ming-Kun and Li, Shao-Yuan and Huang, Sheng-Jun and Chen, Songcan},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={14},
  pages={15438--15446},
  year={2024}
}

1.training

(1) CIFAR80N

Training

• Before training with CECL, initialize training using Promix[https://github.com/Justherozen/ProMix]. Place the obtained correction records and labels into the res_stage1 folder. Then, run:

sh run.sh

• This is a demo under the cifar80N sym20% setting. You can directly use sh run.sh to start training. If you want to customize the training, please go to the config folder and modify the relevant parameters.

2. Requirements

• To install requirements:

pip install -r requirements.txt

• Run in linux (may have some problems in windows)