Enhanced Online Test-time Adaptation with Feature-Weight Cosine Alignment

TL;DR

This paper proposes a cosine alignment method for online test-time adaptation that improves model accuracy and robustness during inference under domain shifts, surpassing existing techniques.

Contribution

It introduces a novel cosine similarity-based optimization approach with a dual-objective loss for more precise predictions and better domain adaptation in OTTA.

Findings

Outperforms state-of-the-art OTTA methods on multiple datasets.

Achieves higher accuracy and robustness against corruptions.

Sets new benchmarks in CIFAR, ImageNet, Office-Home, and DomainNet datasets.

Abstract

Online Test-Time Adaptation (OTTA) has emerged as an effective strategy to handle distributional shifts, allowing on-the-fly adaptation of pre-trained models to new target domains during inference, without the need for source data. We uncovered that the widely studied entropy minimization (EM) method for OTTA, suffers from noisy gradients due to ambiguity near decision boundaries and incorrect low-entropy predictions. To overcome these limitations, this paper introduces a novel cosine alignment optimization approach with a dual-objective loss function that refines the precision of class predictions and adaptability to novel domains. Specifically, our method optimizes the cosine similarity between feature vectors and class weight vectors, enhancing the precision of class predictions and the model's adaptability to novel domains. Our method outperforms state-of-the-art techniques and sets a new benchmark in multiple datasets, including CIFAR-10-C, CIFAR-100-C, ImageNet-C, Office-Home, and DomainNet datasets, demonstrating high accuracy and robustness against diverse corruptions and domain shifts.