In Search of Forgotten Domain Generalization

TL;DR

This paper investigates the true out-of-domain generalization capabilities of models trained on large-scale web datasets, revealing that apparent robustness is often due to in-domain data and proposing new datasets for better evaluation.

Contribution

It introduces large-scale, style-distinct datasets from LAION for rigorous OOD evaluation and systematically explores data mixing strategies to enhance model robustness.

Findings

Models trained on web data perform mainly on in-domain examples.

Optimal data mixing ratios improve cross-domain generalization.

Web-scale training does not inherently guarantee true OOD robustness.

Abstract

Out-of-Domain (OOD) generalization is the ability of a model trained on one or more domains to generalize to unseen domains. In the ImageNet era of computer vision, evaluation sets for measuring a model's OOD performance were designed to be strictly OOD with respect to style. However, the emergence of foundation models and expansive web-scale datasets has obfuscated this evaluation process, as datasets cover a broad range of domains and risk test domain contamination. In search of the forgotten domain generalization, we create large-scale datasets subsampled from LAION -- LAION-Natural and LAION-Rendition -- that are strictly OOD to corresponding ImageNet and DomainNet test sets in terms of style. Training CLIP models on these datasets reveals that a significant portion of their performance is explained by in-domain examples. This indicates that the OOD generalization challenges from the ImageNet era still prevail and that training on web-scale data merely creates the illusion of OOD generalization. Furthermore, through a systematic exploration of combining natural and rendition datasets in varying proportions, we identify optimal mixing ratios for model generalization across these domains. Our datasets and results re-enable meaningful assessment of OOD robustness at scale -- a crucial prerequisite for improving model robustness.