AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Safeguarding Vision-Language Models: Mitigating Vulnerabilities to Gaussian Noise in Perturbation-based Attacks

Jiawei Wang, Yushen Zuo, Yuanjun Chai, Zhendong Liu, Yicheng Fu, Yichun Feng, Kin-Man Lam

ICCV 2025 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Gaussian noise (sigma 0.1) to input imagesInternVL2-8B · MM-Vetperformance score: 59.9 -> 54.4Hurt
add Gaussian noise to input imagesMiniGPT-4-13B · RealToxicityPromptsattack success rate: 34.8 -> 44.1Hurt
add Gaussian noise to input imagesLLaVA-v1.5-7B · RealToxicityPromptsattack success rate: 57.7 -> 60.1Hurt
fine-tune with Robust-VLGuard and noise augmentation instead of VLGuardMiniGPT-4-13B · RealToxicityPrompts (noisy)attack success rate: 43.7 -> 16.5Improved
fine-tune with Robust-VLGuard instead of VLGuardInternVL2-8B · MM-Vet (clean)performance score: 42.9 -> 56.2Improved
fine-tune with Robust-VLGuardLLaVA-v1.5-7B · RealToxicityPrompts (noisy)attack success rate: 60.1 -> 42.3Improved
fine-tune with Robust-VLGuardInternVL2-8B · RealToxicityPrompts (noisy)attack success rate: 57.2 -> 34.5Improved
increase proportion of safety data in instruction ratio (4:1 to 4:4) (ablation)Robust-VLGuard fine-tuning InternVL2-8B · MM-Vet / RealToxicityPromptsattack success rateMixed
increase fine-tuning epochs from 1 to 4 (ablation)Robust-VLGuard fine-tuning InternVL2-8B · MM-Vet / RealToxicityPromptsattack success rateMixed
apply DiffPure (t*=50) to Gaussian-noisy images without noise-augmented fine-tuning (ablation)MiniGPT-4-13B · RealToxicityPromptsattack success rate: 44.1 -> 44.3No clear effect
apply DiffPure (t*=50) to adversarial images (eps=64/255) (ablation)MiniGPT-4-13B · RealToxicityPromptsattack success rate: 67.2 -> 44.5Improved
use DiffPure instead of JailGuard preprocessing (ablation)LLaVA-RobustVLGuard · RealToxicityPrompts (eps=32/255)attack success rate: 48.9 -> 43.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy