Closed alexrame closed 2 years ago
Thank you for letting us know! We've fixed this in commit 0e5d6c. We also checked the history and the conclusion is that it does not affect the results reported in our paper. "CelebA_Blonde" was the original one and later we changed it to "CelebA_Blond" for simplicity during code cleanup after the experiments.
Thank you for the great work and for open-sourcing your benchmark. I believe there is a naming inconsistency for your CelebA experiment.
The dataset is named "CelebA_Blond" here https://github.com/ynysjtu/ood_bench/blob/cac82a5ea97a9d4772066fe40a17d7db185e9615/DomainBed/domainbed/datasets.py#L53 .
In contrast, you define hyperparameter distributions for the "CelebA_Blonde" dataset (with an "e" at the end) here https://github.com/ynysjtu/ood_bench/blob/cac82a5ea97a9d4772066fe40a17d7db185e9615/DomainBed/domainbed/hparams_registry.py#L136
So overall, you may need to fix this. Moreover, could this affect the results reported in your paper? Thanks in advance