Research map / Adversarial robustness and security / Black-box and transfer attacks
Adaptive Image Transformations for Transfer-Based Adversarial Attack
Zheng Yuan, Jie Zhang, Shiguang Shan
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 |
|---|---|---|---|
| adaptive per-image transformation selection (AITL) instead of fixed ADSCM combination | MIFGSM, Incv3 source · defense models, single model | attack success rate: 15.88% average gain | Improved |
| AITL instead of ADSCM, ensemble setting, defense models | MIFGSM ensemble · defense models, ensemble | attack success rate: 5.87% average gain | Improved |
| AITL instead of ADSCM, ensemble setting, normal models | MIFGSM ensemble · normally trained models, ensemble | attack success rate: 1.44% average gain | Improved |
| AITL learned selection instead of Random selection from 12 geometric transformations | Incv3 source · Incv3ens3 | attack success rate: 49.8 -> 69.9 | Improved |
| AITL instead of AutoMA (RL-based policy) | Incv3 source · Incv4 | attack success rate: 91.2 -> 95.8 | Improved |
| AITL-TIM instead of AutoMA-TIM | Incv3 source · Incv3ens3 | attack success rate: 74.8 -> 81.3 | Improved |
| AITL instead of Random on IncResv2ens | Incv3 source · IncResv2ens | attack success rate: 24.5 -> 43.4 | Improved |
| AITL instead of AutoMA-TIM on FD defense | Incv3 source · FD | attack success rate: 84.7 -> 83.8 | No clear effect |
| increase repetitions N from 5 to 10 (ablation) | AITL · defense models Incv3ens3 | attack success rate: 69.9 -> 75.7 | Improved |
| increase repetitions N from 10 to 15 (ablation) | AITL · Incv3ens3 | attack success rate: 75.7 -> 80.4 | Improved |
| increase repetitions N from 5 to 15 (ablation) | AITL · Incv4 (normal model) | attack success rate: 95.8 -> 98.5 | Improved |
| increase transformations per combination M from 2 to 3 (ablation) | AITL · Incv3ens3 | attack success rate: 64.6 -> 67.8 | Improved |
Closest papers by meaning
- Structure Invariant Transformation for better Adversarial Transferability (ICCV 2023 · similarity 0.92)
- Learning to Transform Dynamically for Better Adversarial Transferability (CVPR 2024 · similarity 0.92)
- Improving the Transferability of Adversarial Samples With Adversarial Transformations (CVPR 2021 · similarity 0.92)
- 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 0.89)
- Boosting Adversarial Transferability by Block Shuffle and Rotation (CVPR 2024 · similarity 0.88)
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.
