Research map / Adversarial robustness and security / Black-box and transfer attacks
Distributionally Location-Aware Transferable Adversarial Patches for Facial Images
Xingxing Wei, Shouwei Ruan, Yinpeng Dong, Hang Su, Xiaochun Cao
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 DOP-DTA distribution transfer attack instead of DOP-Rd random sampling (ablation) | DOPatch · LFW face classification SphereFace | attack success rate: 36.6% -> 61.9% | Improved |
| use DOP-LOA location optimization instead of random sampling (ablation) | DOPatch · LFW face classification FaceNet | number of queries: 219 -> 110 | Improved |
| use distribution prior (DOP-DTA) instead of RHDE | RHDE · LFW face classification SphereFace | attack success rate: 31.3% -> 61.9% | Improved |
| use distribution prior (DOP-DTA) instead of RHDE | RHDE · LFW face identification ArcFace | number of queries: 2695 -> 492 | Improved |
| use mean location of distribution (DOP) instead of GDPA white-box location | GDPA · LFW face classification FaceNet | attack success rate: 53.0% -> 72.6% | Improved |
| increase Gaussian components K from 1 to 5 (ablation) | DOPatch query attacks · LFW face classification | attack success rate | Improved |
| increase Gaussian components K beyond 5 (ablation) | DOPatch query attacks · LFW face classification | attack success rate | No clear effect |
| use K=5 components in DOP-DMAT instead of K=1 (ablation) | DOP-DMAT · FaceNet RHDE attack | attack success rate: 12.9% -> 4.5% | Improved |
| use FaceNet as surrogate instead of SphereFace or MobileFaceNet (ablation) | DOPatch · face classification | attack success rate | Improved |
| set entropy regularization lambda to 0.1 (ablation) | DOPatch · FaceNet surrogate | adversarial loss convergence | Improved |
| use meaningful cartoon sticker patterns instead of LaVAN or Gaussian noise (ablation) | DOPatch · face classification | attack success rate | Improved |
| use DOP-DMAT adversarial training instead of standard training | FaceNet · LFW face classification, RHDE attack | attack success rate: 42.2% -> 4.5% | Improved |
Closest papers by meaning
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.87)
- Simultaneously Optimizing Perturbations and Positions for Black-Box Adversarial Patch Attacks (TPAMI 2022 · similarity 0.86)
- False Positives Matter: Multidimensional Localization Evaluation and Training-Free Explainable Adversarial Patch Defense (AAAI 2026 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.84)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.83)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.83)
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.
