PyTorch Implementation of HIERMATCH

Official Code release for HIERMATCH: Leveraging Label Hierarchies for Improving Semi-Supervised Learning
Ashima Garg, Shaurya Bagga, Yashvardhan Singh, Saket Anand.
IEEE Winter Conference on Applications of Computer Vision (WACV 2022)

Citations

If you find this paper useful, please cite our paper:

@InProceedings{Garg_2022_WACV,
    author    = {Garg, Ashima and Bagga, Shaurya and Singh, Yashvardhan and Anand, Saket},
    title     = {HierMatch: Leveraging Label Hierarchies for Improving Semi-Supervised Learning},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2022},
    pages     = {1015-1024}
}

Proposed HIERMATCH

Installation

Clone the repository

$ git clone https://github.com/07Agarg/HIERMATCH
$ cd HIERMATCH

Using the Code

HIERMATCH approach is tested on CIFAR-100 and North American Birds Dataset.

• To run HIERMATCH on CIFAR-100 (for level-2 and level-3)

  • With samples from finest-grained level only and no additional samples from coarser-levels, use the code folder in HIERMATCH-cifar-100/
  • With samples from finest-grained level and partial-labeled samples from coarser-levels, use the code folder in HIERMATCH-cifar-100-partial/.

• To run HIERMATCH on NABirds (for level-2 and level-3)

  • With samples from finest-grained level only and no additional samples from coarser-levels, use the code folder in HIERMATCH-nabirds/
  • With samples from finest-grained level and partial-labeled samples from coarser-levels, use the code folder in HIERMATCH-nabirds-partial/.

Use the command in the respective folders: python train.py

Acknowledgements

The codebase is borrowed from MixMatch

Contact

If you have any suggestion or question, you can leave a message here or contact us directly at ashimag@iiitd.ac.in