SteeringTTA: Guiding Diffusion Trajectories for Robust Test-Time-Adaptation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yu, Jihyun, Oh, Yoojin, Bae, Wonho, Kim, Mingyu, Noh, Junhyug
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918161759076352
author Yu, Jihyun
Oh, Yoojin
Bae, Wonho
Kim, Mingyu
Noh, Junhyug
author_facet Yu, Jihyun
Oh, Yoojin
Bae, Wonho
Kim, Mingyu
Noh, Junhyug
contents Test-time adaptation (TTA) aims to correct performance degradation of deep models under distribution shifts by updating models or inputs using unlabeled test data. Input-only diffusion-based TTA methods improve robustness for classification to corruptions but rely on gradient guidance, limiting exploration and generalization across distortion types. We propose SteeringTTA, an inference-only framework that adapts Feynman-Kac steering to guide diffusion-based input adaptation for classification with rewards driven by pseudo-label. SteeringTTA maintains multiple particle trajectories, steered by a combination of cumulative top-K probabilities and an entropy schedule, to balance exploration and confidence. On ImageNet-C, SteeringTTA consistently outperforms the baseline without any model updates or source data.
format Preprint
id arxiv_https___arxiv_org_abs_2510_14634
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SteeringTTA: Guiding Diffusion Trajectories for Robust Test-Time-Adaptation
Yu, Jihyun
Oh, Yoojin
Bae, Wonho
Kim, Mingyu
Noh, Junhyug
Computer Vision and Pattern Recognition
Test-time adaptation (TTA) aims to correct performance degradation of deep models under distribution shifts by updating models or inputs using unlabeled test data. Input-only diffusion-based TTA methods improve robustness for classification to corruptions but rely on gradient guidance, limiting exploration and generalization across distortion types. We propose SteeringTTA, an inference-only framework that adapts Feynman-Kac steering to guide diffusion-based input adaptation for classification with rewards driven by pseudo-label. SteeringTTA maintains multiple particle trajectories, steered by a combination of cumulative top-K probabilities and an entropy schedule, to balance exploration and confidence. On ImageNet-C, SteeringTTA consistently outperforms the baseline without any model updates or source data.
title SteeringTTA: Guiding Diffusion Trajectories for Robust Test-Time-Adaptation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.14634