Research map / Adversarial robustness and security / Attacks on text and graphs
Towards Highly Transferable Vision-Language Attack via Semantic-Augmented Dynamic Contrastive Interaction
Yuanbo Li, Tianyang Xu, Cong Hu, Tao Zhou, Xiaojun Wu, Josef Kittler
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 |
|---|---|---|---|
| use random negatives instead of lowest-similarity negatives (ablation) | SADCA · Flickr30K, CLIP-CNN source -> ALBEF | TR R@1 ASR: 41.81 -> 45.15 | Improved |
| use random negatives instead of highest-similarity negatives (ablation) | SADCA · Flickr30K, CLIP-CNN source -> ALBEF | TR R@1 ASR: 44.00 -> 45.15 | Improved |
| use semantic augmentation instead of DIM (ablation) | SADCA (DCI) · Flickr30K, ALBEF source -> CLIP-CNN | TR R@1 ASR: 74.46 -> 85.44 | Improved |
| use semantic augmentation instead of SIA (ablation) | SADCA (DCI) · Flickr30K, ALBEF source -> CLIP-CNN | TR R@1 ASR: 73.18 -> 85.44 | Improved |
| use semantic augmentation instead of BSR (ablation) | SADCA (DCI) · Flickr30K, ALBEF source -> CLIP-CNN | TR R@1 ASR: 78.80 -> 85.44 | Improved |
| use semantic augmentation instead of SIA (ablation) | SADCA (DCI) · Flickr30K, ALBEF source -> TCL | TR R@1 ASR: 99.58 -> 98.52 | Hurt |
| add dynamic interaction (versus static interaction with same iterations) (ablation) | SADCA · Flickr30K ITR | ASR | Improved |
| add contrastive interaction, dynamic interaction and semantic augmentation (ablation) | SADCA · Flickr30K ITR | ASR | Improved |
| add larger iterations and SIA to SA-AET | SA-AET · Flickr30K, ALBEF source | avg TR R@1 ASR: 69.74 -> 83.85 | Improved |
| add larger iterations and SIA to SGA | SGA · Flickr30K, ALBEF source | avg TR R@1 ASR: 54.72 -> 67.99 | Improved |
| replace SA-AET(LI)+SIA with SADCA | attack · Flickr30K, ALBEF source -> CLIP-CNN | TR/IR R@1 ASR: 9.19% (TR) and 5.7% (IR) | Improved |
| replace SA-AET(LI)+SIA with SADCA | attack cost · Flickr30K, ALBEF surrogate | run time (h): 11.08 -> 4.40 | Improved |
Closest papers by meaning
- Set-level Guidance Attack: Boosting Adversarial Transferability of Vision-Language Pre-training Models (ICCV 2023 · similarity 0.91)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.90)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.89)
- SLADE: Shielding against Dual Exploits in Large Vision-Language Models (CVPR 2025 · similarity 0.89)
- 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 0.89)
- Mutual-Modality Adversarial Attack with Semantic Perturbation (AAAI 2024 · 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.
