Diversity-aware Buffer for Coping with Temporally Correlated Data Streams in Online Test-time Adaptation

TL;DR

This work tackles online test-time adaptation under temporally correlated, non-i.i.d. data by introducing a Diversity-aware Buffer (DAB) that stores only diverse samples to reduce redundancy and decorrelates consecutive updates. It pairs the buffer with a diversity-and-entropy-based loss weighting scheme and weight ensembling to stabilize learning without relying on batch normalization, enabling robust adaptation across a range of domain shifts. Empirically, DAB achieves state-of-the-art performance on ImageNet-C, ImageNet-R, ImageNet-Sketch, and ImageNet-D109, outperforming strong baselines, especially in correlated (non-i.i.d.) settings; larger buffers further enhance gains, with notable improvements over NOTE and SAR in the reported experiments. The approach advances practical online TTA by handling real-world, temporally structured data streams and offering a principled mechanism to balance diversity and certainty in self-training.

Abstract

Since distribution shifts are likely to occur after a model's deployment and can drastically decrease the model's performance, online test-time adaptation (TTA) continues to update the model during test-time, leveraging the current test data. In real-world scenarios, test data streams are not always independent and identically distributed (i.i.d.). Instead, they are frequently temporally correlated, making them non-i.i.d. Many existing methods struggle to cope with this scenario. In response, we propose a diversity-aware and category-balanced buffer that can simulate an i.i.d. data stream, even in non-i.i.d. scenarios. Combined with a diversity and entropy-weighted entropy loss, we show that a stable adaptation is possible on a wide range of corruptions and natural domain shifts, based on ImageNet. We achieve state-of-the-art results on most considered benchmarks.