https://github.com/MediaBrain-SJTU/DISAM

This repository contains the implementation details for the paper "Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts," accepted at the ICLR 2024.
https://github.com/MediaBrain-SJTU/DISAM

Last synced: 7 months ago
JSON representation

This repository contains the implementation details for the paper "Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts," accepted at the ICLR 2024.

• Host: GitHub
• URL: https://github.com/MediaBrain-SJTU/DISAM
• Owner: MediaBrain-SJTU
• Created: 2023-11-14T04:18:47.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-04-26T10:58:34.000Z (about 1 year ago)
• Last Synced: 2024-05-14T00:16:04.522Z (about 1 year ago)
• Language: Python
• Homepage:
• Size: 10.6 MB
• Stars: 5
• Watchers: 1
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: readme.md

Awesome Lists containing this project

• Awesome-Segment-Anything - [code

README

        
# Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts

This repository contains the implementation details for the paper "Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts," accepted at the International Conference on Learning Representations (ICLR) 2024.

![poster image](./DISAM_poster_online.png)

## Environment Requirements

![Language](https://img.shields.io/badge/language-python-brightgreen)

![Python](https://img.shields.io/badge/Python->=3.9.5-orange)

![PyTorch](https://img.shields.io/badge/PyTorch-=2.0.1-orange)

![NumPy](https://img.shields.io/badge/NumPy->=1.23.5-orange)

## Usage

### Dataset ![repo](https://img.shields.io/badge/repo-DomainBed-informational)

You need to download the dataset on your own and specify the dataset path in the `code/configs/default.py` file. Please refer to [Domainbed repo](https://github.com/facebookresearch/DomainBed).

### Algorithm

The core operations of the algorithm are implemented in the `code/algorithms/DISAM.py` file.

### Example Run Command

```bash

bash ./runs/run_trainer.py --algorithm DISAM_Trainer --dataset pacs --test_domain p --lambda_weight 0.1 --rho 0.05 --lr 1e-3 --batch_size 32 --epoch 50

```

## Citation

If you find our work useful in your research, please consider citing:

```bibtex

@inproceedings{zhang2024domaininspired,

  title={Domain-Inspired Sharpness Aware Minimization Under Domain Shifts},

  author={Ruipeng Zhang and Ziqing Fan and Jiangchao Yao and Ya Zhang and Yanfeng Wang},

  booktitle={The Twelfth International Conference on Learning Representations},

  year={2024},

  url={https://openreview.net/forum?id=I4wB3HA3dJ}

}

```

## License

![License](https://img.shields.io/badge/license-MIT-yellow)

This project is licensed under the MIT License.