Official PyTorch implementation of "Cross-Domain Ensemble Distillation for Domain Generalization" (ECCV 2022)

For more information, please checkout our website and paper.

➡️ We remark that, when with ResNet-18 for a fair comparison, our approach ranks the second and first place in the leaderboard of paperwithcode for PACS and OfficeHome, respectively. (updated at 23.04.02)

conda env create --file environment.yaml
conda activate xded

python pacs_cartoon_train.py  --gpu-id 0 --IPC 16 \
--dataset-config-file configs/datasets/domain_ipc_pacs.yaml \
--config-file configs/xded_default.yaml \
--trainer XDED --remark XDED_UniStyle12 \
MODEL.BACKBONE.NAME resnet18_UniStyle_12

Our code is based on Dassl.pytorch. We thank Kaiyang Zhou for this great repository.

If you have any questions, feedback, or suggestions regarding this repository, feel free to reach out at leekm8474@gmail.com or kyungmoon@postech.ac.kr. I’ll be happy to help!

In case of using this source code for your research, please cite our paper.

@inproceedings{lee2022cross,
  title={Cross-Domain Ensemble Distillation for Domain Generalization},
  author={Lee, Kyungmoon and Kim, Sungyeon and Kwak, Suha},
  booktitle={Proceedings of European Conference on Computer Vision (ECCV)},
  year={2022}
}