Research map / Adversarial robustness and security / Attacks on text and graphs
Segment and Complete: Defending Object Detectors Against Adversarial Patch Attacks With Robust Patch Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SAC's patch segmentation, self adversarial training, and shape completion instead of no defense | Faster R-CNN · COCO 100×100 adaptive attack | mAP (%): 14.4±0.6 -> 44.0±0.3 | Helped |
| Use SAC's patch segmentation, self adversarial training, and shape completion instead of no defense | Faster R-CNN · xView 100×100 adaptive attack | mAP (%): 5.3± 1.1 -> 22.1±0.6 | Helped |
| Use SAC rather than no defense | Faster R-CNN · COCO clean | mAP (%): 49.0 -> 49.0 | No clear effect |
| Use SAC rather than no defense | Faster R-CNN · xView clean | mAP (%): 27.2 -> 27.2 | No clear effect |
| Add self adversarial training to the patch segmenter (ablation) | Patch segmenter on COCO · COCO 100×100 adaptive attack | mAP (%): 18.7±0.3 -> 40.5±0.6 | Helped |
| Add self adversarial training to the patch segmenter (ablation) | Patch segmenter on xView · xView 100×100 adaptive attack | mAP (%): 11.1±0.3 -> 15.4±0.6 | Helped |
| Add shape completion to the self-adversarially trained patch segmenter (ablation) | Patch segmenter on COCO · COCO 100×100 adaptive attack | mAP (%): 40.5±0.6 -> 44.0±0.3 | Helped |
| Add shape completion to the self-adversarially trained patch segmenter (ablation) | Patch segmenter on xView · xView 100×100 adaptive attack | mAP (%): 15.4±0.6 -> 22.1±0.6 | Helped |
| Apply SAC to physical patch attacks rather than leaving the detector undefended | Faster R-CNN on APRICOT · APRICOT development set | targeted attack success rate: 7.97% -> 2.17% | Helped |
| Use JPEG compression rather than no defense | Faster R-CNN · COCO 100×100 adaptive attack | mAP (%): 14.4±0.6 -> 18.0±0.8 | Helped |
| Use adversarial training rather than no defense | Faster R-CNN · COCO clean | mAP (%): 49.0 -> 40.2 | Hurt |
Closest papers by meaning
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.91)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.88)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.87)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.87)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.86)
- GLOW: Global Layout Aware Attacks on Object Detection (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.