Co-learning
Co-learning copied to clipboard
The official implementation of the ACM MM'2021 paper Co-learning: Learning from noisy labels with self-supervision.
Co-learning: Learning from noisy labels with self-supervision
This repository contains a unified framework for co-training-based noisy label learning methods.
The official implementation of the paper Co-learning: Learning from noisy labels with self-supervision is also included.
Introduction
Supported algorithms
- [x] Decoupling (NeurIPS'2017)
- [x] Co-teaching (NeurIPS'2018)
- [x] Co-teaching+ (ICML'2019)
- [x] JoCoR (CVPR'2020)
- [x] Co-learning (MM'2021)
Supported datasets:
- [x] CIFAR-10
- [x] CIFAR-100
Supported synthetic noise types:
- [x] 'sym' (Symmetric noisy labels)
- [x] 'asym' (Asymmetric noisy labels)
- [x] 'ins' (Instance-dependent noisy labels)
Dependency
- numpy
- torch, torchvision
- scipy
- addict
- matplotlib
Citation
If you are interested in our repository and our paper, please cite the following paper:
@inproceedings{tan2021co,
title={Co-learning: Learning from noisy labels with self-supervision},
author={Tan, Cheng and Xia, Jun and Wu, Lirong and Li, Stan Z},
booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
pages={1405--1413},
year={2021}
}