Research map / Adversarial robustness and security / Black-box and transfer attacks
Transform to Transfer: Boosting Adversarial Attack Transferability on Vision-Language Pre-training Models
Yang Li, Jia-Li Yin, Luojun Lin, Wei Lin
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 |
|---|---|---|---|
| add learnable block-level transformation and boosted integrated gradients to SGA-style attack (ablation) | TTA · Flickr30K, ALBEF to CLIP-ViT/CLIP-CNN/TCL | attack success rate | Improved |
| add standard integrated gradients (20 points) in place of standard gradients (ablation) | SGA-style attack · Flickr30K, ALBEF source | attack success rate | Improved |
| add learnable transformation (without boosted IG) (ablation) | TTA · Flickr30K, ALBEF source | attack success rate | Improved |
| increase number of auxiliary transformed images Q_O beyond 30 (ablation) | TTA · Flickr30K, ALBEF source | attack success rate | No clear effect |
| increase block-level transformations L from 1 to 2 (ablation) | TTA · Flickr30K, ALBEF source | attack success rate | Improved |
| increase block-level transformations L beyond 2 (ablation) | TTA · Flickr30K, ALBEF source | TR/IR R@1 and R@5 | Hurt |
| use image-text-image (i-t-i) attack process instead of shorter or longer cycles (ablation) | TTA · Flickr30K, ALBEF source, avg of TCL/CLIP-ViT/CLIP-CNN | attack success rate | Improved |
| replace LSSA with TTA at 20 augmented images | ALBEF source attack · Flickr30K, target CLIP-ViT | TR R@1 attack success rate (%): 53.25 -> 89.08 | Improved |
| add learnable transformations compared with LSSA | TTA · Flickr30K, ALBEF source, 20 augmented images | running time (s): 27.33 -> 28.44 | Hurt |
Closest papers by meaning
- Set-level Guidance Attack: Boosting Adversarial Transferability of Vision-Language Pre-training Models (ICCV 2023 · similarity 0.93)
- Boosting Transferability in Vision-Language Attacks via Diversification along the Intersection Region of Adversarial Trajectory (ECCV 2024 · similarity 0.91)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.91)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.90)
- Learning to Transform Dynamically for Better Adversarial Transferability (CVPR 2024 · similarity 0.90)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 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.
