Code for our paper: "Improving Diversity with Adversarially Learned Transformations for Domain Generalization" To reproduce results for each benchmark, the following steps should be followed.
First, download the data using the following instructions:
- Digits -- data will download automatically if you run
run_alt_mnist.sh
- PACS -- download from https://mega.nz/#F!jBllFAaI!gOXRx97YHx-zorH5wvS6uw
- OfficeHome can be downloaded from the official release https://www.hemanthdv.org/officeHomeDataset.html
- Digits:
bash run_alt_mnist.sh
- PACS:
bash run_alt_pacs.sh
- Office-Home:
bash run_alt_officehome.sh
Part of the code structure is borrowed from RandConv https://github.com/wildphoton/RandConv and Sagnet https://github.com/hyeonseobnam/sagnet . We thank the authors of these papers.