Rethinking the Evaluation Protocol of Domain Generalization
This addresses evaluation fairness in domain generalization for machine learning researchers, though it is incremental as it refines existing protocols rather than introducing new methods.
The paper identifies potential test data information leakage in current domain generalization evaluation protocols, such as supervised pretraining on ImageNet and oracle model selection, and proposes modifications like using self-supervised pretraining or training from scratch and multiple test domains to improve accuracy.
Domain generalization aims to solve the challenge of Out-of-Distribution (OOD) generalization by leveraging common knowledge learned from multiple training domains to generalize to unseen test domains. To accurately evaluate the OOD generalization ability, it is required that test data information is unavailable. However, the current domain generalization protocol may still have potential test data information leakage. This paper examines the risks of test data information leakage from two aspects of the current evaluation protocol: supervised pretraining on ImageNet and oracle model selection. We propose modifications to the current protocol that we should employ self-supervised pretraining or train from scratch instead of employing the current supervised pretraining, and we should use multiple test domains. These would result in a more precise evaluation of OOD generalization ability. We also rerun the algorithms with the modified protocol and introduce new leaderboards to encourage future research in domain generalization with a fairer comparison.