Covariate Shift Detection via Domain Interpolation Sensitivity

Abstract

Covariate shift is a major roadblock in the reliability of image classifers in the real world. Work on covariate shift has been focused on training classifers to adapt or generalize to unseen domains. However, for transparent decision making, it is equally desirable to develop covariate shift detection methods that can indicate whether or not a test image belongs to an unseen domain. In this paper, we introduce a benchmark for covariate shift detection (CSD), that builds upon and complements previous work on domain generalization. We use state-of-the-art OOD detection methods as baselines and fnd them to be worse than simple confdence-based methods on our CSD benchmark. We propose an interpolationbased technique, Domain Interpolation Sensitivity (DIS), based on the simple hypothesis that interpolation between the test input and randomly sampled inputs from the training domain, offers suffcient information to distinguish between the training domain and unseen domains under covariate shift. DIS surpasses all OOD detection baselines for CSD on multiple domain generalization benchmarks.