Rethinking the Evaluation Protocol of Domain Generalization

TL;DR

This paper critically examines current domain generalization evaluation protocols, identifies potential data leakage issues, and proposes modifications such as self-supervised pretraining and multiple test domains for more accurate assessment.

Contribution

It introduces a revised evaluation protocol for domain generalization that reduces test data leakage and provides new leaderboards for fairer comparison of algorithms.

Findings

Current protocols may leak test data information.

Self-supervised pretraining improves evaluation fairness.

New leaderboards facilitate better benchmarking.

Abstract

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.