Research map / Adversarial robustness and security / Attacks on text and graphs
VEAttack: Downstream-agnostic Vision Encoder Attack against Large Vision Language Models
Hefei Mei, Zirui Wang, Shen You, Minjing Dong, Chang Xu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| attack patch token features instead of class token (ablation) | VEAttack · COCO (eps=4/255, LLaVa1.5-7B) | CIDEr after attack (lower better): 25.5 -> 7.1 | Improved |
| also attack class token alongside patch tokens (ablation) | VEAttack · COCO (eps=4/255, LLaVa1.5-7B) | CIDEr after attack (lower better): 7.1 -> 10.5 | Hurt |
| use cosine similarity loss instead of Euclidean (ablation) | VEAttack · COCO (eps=4/255, LLaVa1.5-7B) | CIDEr after attack (lower better): 39.6 -> 7.1 | Improved |
| use cosine similarity instead of K-L divergence (ablation) | VEAttack · VQAv2 (eps=4/255, LLaVa1.5-7B) | VQA accuracy after attack (lower better): 33.2 (K-L) vs 38.4 (cosine) | Mixed |
| reduce attack steps from 100 to 10 (ablation) | VEAttack · COCO and TextVQA | performance degradation: 84.4% reduction at t=10 | No clear effect |
| increase perturbation budget beyond 8/255 (ablation) | VEAttack · COCO and TextVQA | performance degradation | No clear effect |
| use a robust (FARE) vision encoder as the victim | LLaVa1.5-7B · COCO transfer | CIDEr after attack: 3.7 -> 48.6 (FARE self-attack) | Improved |
| craft adversarial examples on a robust encoder (FARE) instead of CLIP | VEAttack transfer · COCO transfer to CLIP-based LLaVa1.5-7B | CIDEr after attack: CLIP source: 3.7 own, FARE source: 38.7 on CLIP target | Improved |
| switch from white-box Ensemble to VEAttack (eps=8/255, t=50) | attack pipeline · COCO full dataset | attack time (h): 41.9 -> 5.3 | Improved |
Later papers that compared with this method
1 later accepted paper reports VEAttack as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- PA-Attack: Guiding Gray-Box Attacks on LVLM Vision Encoders with Prototypes and Attention (CVPR 2026 · better on 6 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PA-Attack: Guiding Gray-Box Attacks on LVLM Vision Encoders with Prototypes and Attention (CVPR 2026 · similarity 0.93)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.90)
- A Frustratingly Simple Yet Highly Effective Attack Baseline: Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1 (NeurIPS 2025 · similarity 0.90)
- SLADE: Shielding against Dual Exploits in Large Vision-Language Models (CVPR 2025 · similarity 0.90)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.89)
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.
