Research map / Adversarial robustness and security / Black-box and transfer attacks
Simultaneously Optimizing Perturbations and Positions for Black-Box Adversarial Patch Attacks
Xingxing Wei, Ying Guo, Jie Yu, Bo Zhang
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 |
|---|---|---|---|
| optimize position and perturbation simultaneously instead of alternately (LO) | adversarial patch attack · Face dodging attack on FaceNet | ASR: 76.82% -> 96.65% | Improved |
| optimize simultaneously instead of two-step PatchAttack | adversarial patch attack · Face impersonation attack on FaceNet | ASR: 68.95% -> 72.83% | Improved |
| optimize simultaneously instead of two-step PatchAttack | adversarial patch attack · Face impersonation attack on CosFace50 | ASR: 35.56% -> 40.08% | Improved |
| use ensemble surrogate instead of single ArcFace50 surrogate (ablation) | our method · Face dodging attack on FaceNet | ASR: 89.37% -> 96.65% | Improved |
| add weight parameter and step size to position-only learning (ablation) | our method · Face dodging attack (average) | ASR: 69.80% (position only), further improved by weights and step size | Improved |
| initialize ZO-AdaMM with SO patch instead of random initialization (ablation) | ZO attack · Face dodging attack on FaceNet | ASR: 65.79% -> 97.73% | Improved |
| initialize ZO attack with SO patch instead of random initialization (ablation) | ZO attack · Face impersonation attack on FaceNet | ASR: 50.52% -> 87.37% | Improved |
| apply DW preprocessing defense | our attack · Face dodging attack on FaceNet | ASR: 96.65% -> 81.71% | Hurt |
| apply LGS preprocessing defense | our attack · Face dodging attack on FaceNet | ASR: drops by 1.01% | No clear effect |
| apply adversarial training to target model | GDPA attack · Face dodging attack on FaceNet | ASR: down 48.51% | Hurt |
| apply adversarial training to target model | our attack · Face dodging attack on FaceNet | ASR: down 2.69% | No clear effect |
| translate patch by 5 pixels | our attack · Face dodging attack on FaceNet | ASR: 96.65% -> 79.24% | Hurt |
Closest papers by meaning
- REAP: A Large-Scale Realistic Adversarial Patch Benchmark (ICCV 2023 · similarity 0.88)
- Breaking Rigidity in Adversarial Patch Attacks (ECCV 2026 · similarity 0.88)
- AdvDiffusion: Adversarial Patches Generation for Face Recognition With High Transferability in Physical Domain (TPAMI 2026 · similarity 0.87)
- Adaptive Perturbation for Adversarial Attack (TPAMI 2024 · similarity 0.87)
- IAP: Invisible Adversarial Patch Attack through Perceptibility-Aware Localization and Perturbation Optimization (ICCV 2025 · similarity 0.87)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · 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.
