Code release for Open Domain Generalization with Domain-Augmented Meta-Learning (CVPR2021)
- Here are some datasets you may need. PACS and Office-Home for the open domain generalization experiments on them. Office-31, STL-10, Visda2017 and DomainNet for the Multi-Datasets experiment.
Dataset | Link |
---|---|
PACS | https://dali-dl.github.io/project_iccv2017.html |
Office-Home | https://www.hemanthdv.org/officeHomeDataset.html |
Office-31 | http://www.eecs.berkeley.edu/~mfritz/domainadaptation/ |
STL-10 | https://cs.stanford.edu/~acoates/stl10/ |
Visda2017 | http://ai.bu.edu/visda-2017/ |
DomainNet | http://ai.bu.edu/M3SDA/ |
- We provide the labels and train-val-test splits for these datasets in
data/
folder.
- Python 3.8
- PyTorch 1.5.0
- Download the DATASET you need. Move the
image_list
folder of the DATASET (which we provide indata/DATASET/
) to the directory of the DATASET. - We provide scripts in
src/scripts/
. Complete the configuration of experiments, such as the path to the DATASET, thenbash run_train.sh
for training on source domains and testing on target domain data from known classes. - After training and saving the model checkpoints,
bash run_validate.sh
for testing on the whole target domain, including both known and unknown classes.
If you find this code or our paper useful, please consider citing:
@inproceedings{shu2021open,
title={Open Domain Generalization with Domain-Augmented Meta-Learning},
author={Shu, Yang and Cao, Zhangjie and Wang, Chenyu and Wang, Jianmin and Long, Mingsheng},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={9624--9633},
year={2021}
}
If you have any problems about our code, feel free to contact