DICS: Find Domain-Invariant and Class-Specific Features for Out-of-Distribution Generalization

TL;DR

The paper introduces DICS, a novel method that extracts domain-invariant and class-specific features to improve out-of-distribution generalization in vision tasks, addressing confounders and spurious correlations.

Contribution

DICS is the first approach to simultaneously learn domain-invariant and class-specific features, enhancing OOD performance by mitigating confounders and spurious correlations.

Findings

DICS outperforms existing methods on standard benchmarks.

Visualizations show DICS effectively identifies key class features.

DICS improves class discrimination and domain invariance in OOD scenarios.

Abstract

While deep neural networks have made remarkable progress in various vision tasks, their performance typically deteriorates when tested in out-of-distribution (OOD) scenarios. Many OOD methods focus on extracting domain-invariant features but neglect whether these features are unique to each class. Even if some features are domain-invariant, they cannot serve as key classification criteria if shared across different classes. In OOD tasks, both domain-related and class-shared features act as confounders that hinder generalization. In this paper, we propose a DICS model to extract Domain-Invariant and Class-Specific features, including Domain Invariance Testing (DIT) and Class Specificity Testing (CST), which mitigate the effects of spurious correlations introduced by confounders. DIT learns domain-related features of each source domain and removes them from inputs to isolate domain-invariant class-related features. DIT ensures domain invariance by aligning same-class features across different domains. Then, CST calculates soft labels for those features by comparing them with features learned in previous steps. We optimize the cross-entropy between the soft labels and their true labels, which enhances same-class similarity and different-class distinctiveness, thereby reinforcing class specificity. Extensive experiments on widely-used benchmarks demonstrate the effectiveness of our proposed algorithm. Additional visualizations further demonstrate that DICS effectively identifies the key features of each class in target domains.