NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation

TL;DR

This paper introduces a robust test-time adaptation method for non-i.i.d. data streams, using instance-aware normalization and class-balanced sampling, outperforming existing methods in correlated test scenarios.

Contribution

The paper proposes a novel TTA scheme that is robust to temporal correlations in test data, addressing a key limitation of prior i.i.d.-assumed methods.

Findings

Outperforms state-of-the-art TTA methods on non-i.i.d. data streams.

Achieves comparable performance to i.i.d.-based algorithms under correlated data.

Demonstrates effectiveness on real-world non-i.i.d. datasets.

Abstract

Test-time adaptation (TTA) is an emerging paradigm that addresses distributional shifts between training and testing phases without additional data acquisition or labeling cost; only unlabeled test data streams are used for continual model adaptation. Previous TTA schemes assume that the test samples are independent and identically distributed (i.i.d.), even though they are often temporally correlated (non-i.i.d.) in application scenarios, e.g., autonomous driving. We discover that most existing TTA methods fail dramatically under such scenarios. Motivated by this, we present a new test-time adaptation scheme that is robust against non-i.i.d. test data streams. Our novelty is mainly two-fold: (a) Instance-Aware Batch Normalization (IABN) that corrects normalization for out-of-distribution samples, and (b) Prediction-balanced Reservoir Sampling (PBRS) that simulates i.i.d. data stream from non-i.i.d. stream in a class-balanced manner. Our evaluation with various datasets, including real-world non-i.i.d. streams, demonstrates that the proposed robust TTA not only outperforms state-of-the-art TTA algorithms in the non-i.i.d. setting, but also achieves comparable performance to those algorithms under the i.i.d. assumption. Code is available at https://github.com/TaesikGong/NOTE.