megvii-research / FullMatch

Official implementation of FullMatch (CVPR2023)

FullMatch

Boosting Semi-supervised Learning by Exploiting All Unlabeled Data

Yuhao Chen, Xin Tan, Borui Zhao, Zhaowei Chen, Renjie Song, Jiajun Liang, Xuequan Lu

CVPR 2023, Arxiv

This repo is the Megengine implementation of FullMatch. The Pytorch implementation will be released soon.

Experiment

1. Install MegEngine (version==1.12.2/1.12.3)

2. For training FullMatch:

python fullmatch.py --c config/fullmatch/fullmatch_cifar100.yaml

3. For training FullFlex:

python fullflex.py --c config/fullflex/fullflex_cifar100.yaml

Note

Since the official Megengine does not support many classification benchmarks (e.g., SVHN, STL10), we will release them in the Pytorch implementation.

We thanks the TorchSSL project for reference.

Log & Models

All origin train logs and models are in this link

Liscense

FullMatch is released under the Apache 2.0 license. See LICENSE for details.

Citation

@inproceedings{chen2023boosting,
  title={Boosting Semi-Supervised Learning by Exploiting All Unlabeled Data},
  author={Chen, Yuhao and Tan, Xin and Zhao, Borui and Chen, Zhaowei and Song, Renjie and Liang, Jiajun and Lu, Xuequan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7548--7557},
  year={2023}
}