Research map / Adversarial robustness and security / Attacks on text and graphs
AntiStyler: Defending Object Detection Models Against Adversarial Patch Attacks Using Style Removal
Idan Yankelev, Edita Grolman, Yarin Yerushalmi Levi, Amit Giloni, Omer Hofman, Toshiya Shimizu, Yuval Elovici, Asaf Shabtai
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 |
|---|---|---|---|
| remove random padding (ablation) | AntiStyler · COCO, Google patch (benign images) | mAP@0.5: 51.6 -> 38.8 | Hurt |
| add random-value padding (ablation) | AntiStyler · COCO, M-PGD patch (adversarial images) | mAP@0.5: 35.7 -> 38.0 | Improved |
| add filter, enhancement and mask phases (ablation) | AntiStyler · COCO, Google patch (adversarial images) | mAP@0.5: 29.3 -> 32.5 | Improved |
| add enhancement phase (spatial filters) to raw mask (ablation) | AntiStyler · COCO, M-PGD patch (adversarial images) | mAP@0.5: 35.3 -> 38.0 | Improved |
| VGG19 backbone instead of ResNet50 (ablation) | AntiStyler · COCO, M-PGD patch (adversarial images) | mAP@0.5: 32.4 -> 38.0 | Improved |
| VGG19 backbone instead of EfficientNetV2 (ablation) | AntiStyler · COCO, Google patch (adversarial images) | mAP@0.5: 24.2 -> 32.5 | Improved |
| apply AntiStyler against black-box adaptive patch | AntiStyler · COCO adaptive attacks | mAP@0.5: 23.4 -> 30.5 | Improved |
| apply AntiStyler against white-box adaptive patch | AntiStyler · COCO adaptive attacks | mAP@0.5: 18.7 -> 23.9 | Improved |
| AntiStyler vs PAD on printable patch P2 | AntiStyler · INRIA, natural patch P2 | mAP@0.5: 74.3 -> 74.4 | No clear effect |
| AntiStyler vs KDAT on DETR DPatch | AntiStyler · COCO DPatch, DETR | mAP@0.5: 45.5 -> 44.4 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- KDAT — KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models (AAAI 2025) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Segment and Complete: Defending Object Detectors Against Adversarial Patch Attacks With Robust Patch Detection (CVPR 2022 · similarity 0.91)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.87)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
- Can't Slow Me Down: Learning Robust and Hardware-Adaptive Object Detectors against Latency Attacks for Edge Devices (CVPR 2025 · similarity 0.85)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.85)
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.
