Test Time Adaptation Using Adaptive Quantile Recalibration

TL;DR

This paper introduces Adaptive Quantile Recalibration (AQR), a novel test-time adaptation method that aligns activation distribution quantiles to improve model robustness across diverse and changing data domains without retraining.

Contribution

AQR is the first method to adapt pre-activation distributions by quantile alignment across various normalization layers, capturing full distribution shapes and improving stability with tail calibration.

Findings

AQR outperforms existing methods on CIFAR-10-C, CIFAR-100-C, and ImageNet-C benchmarks.

AQR effectively generalizes across architectures using BatchNorm, GroupNorm, or LayerNorm.

AQR provides robust unsupervised adaptation without retraining models.

Abstract

Domain adaptation is a key strategy for enhancing the generalizability of deep learning models in real-world scenarios, where test distributions often diverge significantly from the training domain. However, conventional approaches typically rely on prior knowledge of the target domain or require model retraining, limiting their practicality in dynamic or resource-constrained environments. Recent test-time adaptation methods based on batch normalization statistic updates allow for unsupervised adaptation, but they often fail to capture complex activation distributions and are constrained to specific normalization layers. We propose Adaptive Quantile Recalibration (AQR), a test-time adaptation technique that modifies pre-activation distributions by aligning quantiles on a channel-wise basis. AQR captures the full shape of activation distributions and generalizes across architectures employing BatchNorm, GroupNorm, or LayerNorm. To address the challenge of estimating distribution tails under varying batch sizes, AQR incorporates a robust tail calibration strategy that improves stability and precision. Our method leverages source-domain statistics computed at training time, enabling unsupervised adaptation without retraining models. Experiments on CIFAR-10-C, CIFAR-100-C, and ImageNet-C across multiple architectures demonstrate that AQR achieves robust adaptation across diverse settings, outperforming existing test-time adaptation baselines. These results highlight AQR's potential for deployment in real-world scenarios with dynamic and unpredictable data distributions.