Domain Generalization via Gradient Surgery

This repository contains the source code corresponding to the paper "Domain Generalization via Gradient Surgery" (ICCV 2021). You can check out our paper here: https://arxiv.org/abs/2108.01621.

Instructions

This project uses Python 3.8.10 and PyTorch 1.10.0.

Data:

1. Download the PACS (Li et al., 2017), VLCS (Fang et al., 2013) and Office-Home (Venkateswara et al., 2017) datasets and put them in data/raw/.
2. Resize images and generate training, validation and test splits. Run ./00_prepare_data.sh after installing the project environment (instructions below).

Project environment:

1. Create and activate virtual environment: 1) python3 -m venv env, 2) source env/bin/activate
2. Install required packages: pip install -r requirements.txt
3. Install project modules (src): pip install -e .

Simulations:

To run simulations across all datasets (PACS, VLCS and Office-Home) and methods (Deep-All, Agr-Sum, Agr-Rand and PCGrad), execute ./01_run_trials.sh.

If you want to run a particular combination of dataset and method, use the train_model.py script. For example, the following instruction:

python scripts/train_model.py \
    --data_dir=data/processed \
    --results_dir=results/train \
    --dataset=PACS \
    --method=deep-all

will run Deep-All on PACS and save the results in results/train.

Reference

• Mansilla, L., Echeveste, R., Milone, D. H., & Ferrante, E. (2021). Domain generalization via gradient surgery. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 6630-6638).

License

MIT