Research map / Adversarial robustness and security / Attacks on text and graphs
Naturalistic Physical Adversarial Patch for Object Detectors
Yu-Chih-Tuan Hu, Bo-Han Kung, Daniel Stanley Tan, Jun-Cheng Chen, Kai-Lung Hua, Wen-Huang Cheng
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 |
|---|---|---|---|
| train with in-plane rotation (ablation) | Ours-YOLOv4tiny · INRIA, rotation test | mAP (%): 13.42 -> 10.16 | Improved |
| train with all transformations (ablation) | Ours-YOLOv4tiny · INRIA, all-transformations test | mAP (%): 63.37 -> 60.44 | Hurt |
| train with random translation (ablation) | Ours-YOLOv4tiny · INRIA, translation test | mAP (%): 30.15 -> 31.63 | Improved |
| train with random occlusion (ablation) | Ours-YOLOv4tiny · INRIA, occlusion test | mAP (%): 34.92 -> 37.25 | Mixed |
| increase patch scale from 0.05 to 0.3 (ablation) | Ours-YOLOv4tiny · INRIA 0.0-1.0 | AP (%): 95.04 -> 1.64 | Improved |
| larger-pedestrian subset vs small (patch scale 0.1) (ablation) | Ours-YOLOv4tiny · INRIA 0.2-1.0 vs 0.0-0.2 | AP (%): 75.48 -> 55.17 | Mixed |
| train on INRIA and MPII mix instead of INRIA only (ablation) | Ours-YOLOv4tiny · INRIA test | mAP (%): 8.67 -> 2.69 | Hurt |
| replace BigGAN with StyleGAN2 generator (ablation) | Ours-YOLOv4tiny · INRIA | mAP (%): 8.67 -> 34.34 | Mixed |
| use GAN latent-space patch vs Adversarial Patches pixel-space baseline | Ours-YOLOv4tiny · INRIA, victim YOLOv4tiny | mAP (%): 3.25 -> 8.67 | Hurt |
| use class dog patch vs bulbul patch (ablation) | Ours-YOLOv4tiny · INRIA, victim YOLOv4tiny | mAP (%): 8.67 -> 24.36 | Mixed |
Closest papers by meaning
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.91)
- APRICOT: A Dataset of Physical Adversarial Attacks on Object Detection (ECCV 2020 · similarity 0.91)
- Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights (ICCV 2025 · similarity 0.91)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.90)
- Unified Adversarial Patch for Cross-Modal Attacks in the Physical World (ICCV 2023 · similarity 0.90)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 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.
