ICML‘2021: Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels (PyTorch implementation).

This is the code for the paper: Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels.
Songhua Wu^*, Xiaobo Xia^*, Tongliang Liu, Bo Han, Mingming Gong, Nannan Wang, Haifeng Liu, Gang Niu.

If you find this code useful for your research, please cite

@inproceedings{wu2021class2simi,
  title={Class2simi: A noise reduction perspective on learning with noisy labels},
  author={Wu, Songhua and Xia, Xiaobo and Liu, Tongliang and Han, Bo and Gong, Mingming and Wang, Nannan and Liu, Haifeng and Niu, Gang},
  booktitle={International Conference on Machine Learning},
  pages={11285--11295},
  year={2021},
  organization={PMLR}
}

We implement our methods by PyTorch on Nvidia GeForce RTX 3090 Ti. The environment is as bellow:

• Ubuntu 20.04 Desktop
• PyTorch, version >= 0.4.1
• CUDA, version >= 9.0
• Anaconda3

We process the raw images and labels into .npy format. The MNIST dataset can be found in the /data folder of this repository. The CIFAR10 dataset can be download here.

Here is an example:

python main.py --dataset mnist --noise_type s --r 0.2 --loss forward