R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sheng, Lijun, Liang, Jian, Wang, Zilei, He, Ran
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908504955027456
author Sheng, Lijun
Liang, Jian
Wang, Zilei
He, Ran
author_facet Sheng, Lijun
Liang, Jian
Wang, Zilei
He, Ran
contents Vision-language models (VLMs), such as CLIP, have gained significant popularity as foundation models, with numerous fine-tuning methods developed to enhance performance on downstream tasks. However, due to their inherent vulnerability and the common practice of selecting from a limited set of open-source models, VLMs suffer from a higher risk of adversarial attacks than traditional vision models. Existing defense techniques typically rely on adversarial fine-tuning during training, which requires labeled data and lacks of flexibility for downstream tasks. To address these limitations, we propose robust test-time prompt tuning (R-TPT), which mitigates the impact of adversarial attacks during the inference stage. We first reformulate the classic marginal entropy objective by eliminating the term that introduces conflicts under adversarial conditions, retaining only the pointwise entropy minimization. Furthermore, we introduce a plug-and-play reliability-based weighted ensembling strategy, which aggregates useful information from reliable augmented views to strengthen the defense. R-TPT enhances defense against adversarial attacks without requiring labeled training data while offering high flexibility for inference tasks. Extensive experiments on widely used benchmarks with various attacks demonstrate the effectiveness of R-TPT. The code is available in https://github.com/TomSheng21/R-TPT.
format Preprint
id arxiv_https___arxiv_org_abs_2504_11195
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning
Sheng, Lijun
Liang, Jian
Wang, Zilei
He, Ran
Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
Vision-language models (VLMs), such as CLIP, have gained significant popularity as foundation models, with numerous fine-tuning methods developed to enhance performance on downstream tasks. However, due to their inherent vulnerability and the common practice of selecting from a limited set of open-source models, VLMs suffer from a higher risk of adversarial attacks than traditional vision models. Existing defense techniques typically rely on adversarial fine-tuning during training, which requires labeled data and lacks of flexibility for downstream tasks. To address these limitations, we propose robust test-time prompt tuning (R-TPT), which mitigates the impact of adversarial attacks during the inference stage. We first reformulate the classic marginal entropy objective by eliminating the term that introduces conflicts under adversarial conditions, retaining only the pointwise entropy minimization. Furthermore, we introduce a plug-and-play reliability-based weighted ensembling strategy, which aggregates useful information from reliable augmented views to strengthen the defense. R-TPT enhances defense against adversarial attacks without requiring labeled training data while offering high flexibility for inference tasks. Extensive experiments on widely used benchmarks with various attacks demonstrate the effectiveness of R-TPT. The code is available in https://github.com/TomSheng21/R-TPT.
title R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning
topic Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.11195