Lune

CVPR Findings2026

Reliable Test-time Adaptation Via Evidential Uncertainty Modeling in Vision–Language Models

Yiwei You, Zan Chen, Bo Wang, Xiaofei Zhou

Abstract

Test-time Prompt Tuning (TPT) has proven effective for adapting vision-language models such as CLIP to unseen domains without retraining. However, existing TPT approaches primarily depend on first-order prediction entropy to select confident augmentations, limiting their exploitation of diverse and informative views. In this paper, we propose Reliable Test-time Adaptation (RTA), an uncertainty-aware framework that models second-order uncertainty to enable more reliable and efficient adaptation. RTA reinterprets CLIP's cosine similarities as evidential quantities and employs Evidential Deep Learning (EDL) to construct a Dirichlet-based opinion model that jointly captures both model confidence and epistemic uncertainty. This formulation supports the principled identification and aggregation of reliable augmented views beyond conventional entropy-based filtering. Moreover, we propose a novel Reliable Evidential Entropy (REE) loss to stabilize adaptation and an uncertainty-aware cache model for continual refinement during inference. Extensive experiments on out-ofdistribution and cross-domain benchmark demonstrate that RTA consistently improves both accuracy and robustness over existing test-time adaptation methods, without requiring additional train set or supervision.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models cites Conditional Prompt Learning for Vision-Language Models
  • Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models cites AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
  • Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models cites The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization
  • Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models cites Learning Transferable Visual Models From Natural Language Supervision
  • Conditional Prompt Learning for Vision-Language Models cites The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization
  • Conditional Prompt Learning for Vision-Language Models cites Learning Transferable Visual Models From Natural Language Supervision
  • The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization cites AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
  • Learning Transferable Visual Models From Natural Language Supervision cites The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on26

  1. Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
  2. The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution GeneralizationDan Hendrycks, Steven Basart, Norman Mu, Saurav Kadavath et al.ICCV 2021 · 2,294 citations
  3. AugMix: A Simple Data Processing Method to Improve Robustness and UncertaintyDan Hendrycks, Norman Mu, Ekin Dogus Cubuk, Barret Zoph et al.ICLR 2020 · 1,572 citations
  4. Conditional Prompt Learning for Vision-Language ModelsKaiyang Zhou, Jingkang Yang, Chen Change Loy, Ziwei LiuCVPR 2022 · 1,438 citations
  5. Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language ModelsManli Shu, Weili Nie, De-An Huang, Zhiding Yu et al.NeurIPS 2022 · 603 citations

Related papers

  1. R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt TuningLijun Sheng, Jian Liang, Zilei Wang, Ran HeCVPR 2025
  2. Diverse Data Augmentation with Diffusions for Effective Test-time Prompt TuningChun-Mei Feng, Kai Yu, Yong Liu, Salman Khan et al.ICCV 2023 · 172 citations
  3. Hierarchical Knowledge Prompt Tuning for Multi-task Test-Time AdaptationQiang Zhang, Mengsheng Zhao, Jiawei Liu, Fanrui Zhang et al.CVPR 2025
  4. Adaptive Debiasing Tsallis Entropy for Test-Time AdaptationXiangyu Wu, Dongming Jiang, Feng Yu, Yueying Tian et al.ICLR 2026 · 2 citations
  5. CLIPTTA: Robust Contrastive Vision-Language Test-Time AdaptationMarc Lafon, Gustavo Adolfo Vargas Hakim, Clément Rambour, Christian Desrosiers et al.NeurIPS 2025 · 5 citations
Dusk over the sea between two cliffs drawn in fine vertical lines