Awesome
CoMatch: Semi-supervised Learning with Contrastive Graph Regularization, ICCV 2021 (Salesforce Research).
<img src="comatch.gif" width="600">This is a PyTorch implementation of the <a href="https://arxiv.org/abs/2011.11183">CoMatch paper</a> <a href="https://blog.einstein.ai/comatch-advancing-semi-supervised-learning-with-contrastive-graph-regularization/">[Blog]</a>:
<pre> @inproceedings{CoMatch, title={Semi-supervised Learning with Contrastive Graph Regularization}, author={Junnan Li and Caiming Xiong and Steven C.H. Hoi}, booktitle={ICCV}, year={2021} }</pre>Requirements:
- PyTorch ≥ 1.4
- pip install tensorboard_logger
- download and extract <a href="https://www.cs.toronto.edu/~kriz/cifar.html">cifar-10 dataset</a> into ./data/
To perform semi-supervised learning on CIFAR-10 with 4 labels per class, run:
<pre>python Train_CoMatch.py --n-labeled 40 --seed 1 </pre>The results using different random seeds are:
| seed | 1 | 2 | 3 | 4 | 5 | avg |
|---|---|---|---|---|---|---|
| accuracy | 93.71 | 94.10 | 92.93 | 90.73 | 93.97 | 93.09 |
ImageNet
For ImageNet experiments, see ./imagenet/