MixTTA: Low-Rank Cross-Channel Mixing for Reliable Test-Time Adaptation

2026-06-26

AI Summary

This work addresses the limitations of existing test-time adaptation methods, which rely on per-channel affine parameters in normalization layers and struggle to handle cross-channel structural shifts caused by distributional shifts. To overcome this, the paper proposes MixTTA—a lightweight plug-in module that enhances model robustness by enabling intra-layer channel interaction through a low-rank cross-channel mixing mechanism integrated into normalization layers. MixTTA innovatively combines decoupled projection, which strictly separates the new pathway from the conventional diagonal affine transformation, and spectral projection, which prevents rank collapse under non-stationary test streams. The method seamlessly integrates into any normalization-based test-time adaptation framework, consistently outperforming strong baselines across both standard and real-world scenarios, and effectively mitigating adaptation failures under complex conditions.

Abstract

Test-Time Adaptation (TTA) methods commonly update the affine parameters of normalization layers to adapt deployed models under distribution shifts. However, per-channel affine parameters perform axis-aligned scaling and shifting, making them geometrically incapable of correcting cross-channel structural changes induced by distribution shift. To address this limitation, we propose MixTTA, a lightweight plug-in module that equips normalization layers with a low-rank cross-channel transformation, enabling inter-channel mixing at each layer. To ensure that the low-rank branch captures only cross-channel interactions, we also propose Decoupling Projection that enforces strict separation from the diagonal affine path, along with Spectral Projection that prevents rank-1 collapse under non-stationary test streams. MixTTA can be seamlessly integrated into any existing normalization-based TTA method. Experiments in both standard and wild TTA settings show consistent improvements over strong baselines while mitigating adaptation failure under challenging conditions. The source code is publicly available at https://github.com/delta6189/MixTTA.

Problem

• Test-Time Adaptation
• distribution shift
• cross-channel interaction
• normalization layers
• adaptation failure

Innovation

• Test-Time Adaptation
• Low-Rank Transformation
• Cross-Channel Mixing
• Normalization Layers
• Distribution Shift