Research map / Adversarial robustness and security / Black-box and transfer attacks
SMP-Attack: Boosting the Transferability of Feature Importance-based Adversarial Attack with Semantics-aware Multi-granularity Patchout
Wen Yang, Guodong Liu, Di Ming
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 SMP-Attack instead of MFAA | feature importance attack · 6 undefended CNNs, 3 surrogates avg | attack success rate (%): 92.03% outperforming MFAA by 3.77% | Improved |
| combine SMP with PIM and DIM instead of MFAA | PIDI attacks · 6 undefended CNNs | attack success rate (%): 93.93%, surpassing PIDI-MFAA by 1.07% | Improved |
| use SMP-Attack instead of MFAA | feature importance attack · 4 defended CNNs | attack success rate (%): 3.93% improvement over MFAA | Improved |
| combine SMP with PIM and DIM instead of MFAA | PIDI attacks · 4 defended CNNs | attack success rate (%): 1.60% compared to PIDI-MFAA | Improved |
| use SMP-Attack instead of MFAA | feature importance attack · 4 ViTs | attack success rate (%): approximately 6.47% compared to MFAA | Improved |
| combine SMP with PIM and DIM instead of MFAA | PIDI attacks · 4 ViTs | attack success rate (%): increase of 5.47% over PIDI-MFAA | Improved |
| add multi-granularity (vary both d and c) instead of single granularity (ablation) | SMP patchout · ImageNet-compatible black-box | attack success rate (%) | Improved |
| add multi-stage training instead of joint or alternating single-stage multi-layer training (ablation) | SMP-Attack · ImageNet-compatible black-box | attack success rate (%) | Improved |
| use large irregular patches D=[3,4,5] with CSharp in the intermediate layer (ablation) | single-stage SMP · Vgg-16 surrogate, Inc-v3/IncRes-v2ens/Swin-B | attack success rate (%) | Improved |
| use small regular patches D=[1,2,3] with CSmooth in the shallow layer (ablation) | multi-stage SMP · Vgg-16 surrogate, Inc-v3/IncRes-v2ens/Swin-B | attack success rate (%) | Improved |
| set modify probability pm between 0.2 and 0.3 (ablation) | SMP-Attack · Res152-v1 surrogate, Vgg-16/Inc-v3ens3/PiT-S | attack success rate (%) | Improved |
| set ensemble number M to 60 (ablation) | SMP-Attack · Res152-v1 surrogate, Vgg-16/Inc-v3ens3/PiT-S | attack success rate (%) | Improved |
Closest papers by meaning
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.87)
- Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024 · similarity 0.87)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · similarity 0.87)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.86)
- Improving Transferable Targeted Attacks with Feature Tuning Mixup (CVPR 2025 · similarity 0.86)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.86)
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.
