Research map / Adversarial robustness and security / Attacks on text and graphs
Jedi: Entropy-Based Localization and Removal of Adversarial Patches
Bilel Tarchoun, Anouar Ben Khalifa, Mohamed Ali Mahjoub, Nael Abu-Ghazaleh, Ihsen Alouani
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 entropy heat maps and autoencoder mask reconstruction to localize patches | Jedi · ImageNet + [cite] | patch localization IoU > 0.5: 87.20% vs 44.50% (LGS) and 10.85% (Jujutsu) | Improved |
| Use entropy heat maps and autoencoder mask reconstruction to localize patches | Jedi · Pascal VOC 07 + [cite] | patch localization IoU > 0.5: 90.71% vs 34.30% (LGS) and 19.22% (Jujutsu) | Improved |
| Use entropy heat maps and autoencoder mask reconstruction to localize patches | Jedi · CASIA + [cite] | patch localization IoU > 0.5: 93.49% vs 57.25% (LGS) | Improved |
| Use entropy heat maps and autoencoder mask reconstruction to localize patches | Jedi · INRIA + [cite] | patch localization IoU > 0.5: 38.80% vs 73,11% (LGS) | Hurt |
| Apply Jedi entropy localization and inpainting | Jedi with ResNet50 · ImageNet | robust accuracy: 39.26% -> 64.34% | Improved |
| Apply Jedi entropy localization and inpainting | Jedi with ResNet50 · Pascal VOC 07 | robust accuracy: 26.94% -> 66.40% | Improved |
| Apply Jedi entropy localization and inpainting | Jedi with YOLO · CASIA | robust average precision: 39.60% -> 88.21% | Improved |
| Apply Jedi entropy localization and inpainting | Jedi with YOLO · INRIA | robust average precision: 12.17% -> 28.03% | Improved |
| Replace regular patch generation with entropy-bounded low-entropy patch generation (ablation) | Adversarial patch attack on Pascal VOC 07 · Pascal VOC 07 | attack success rate: 64% -> 8% | Hurt |
| Lower the patch entropy budget (ablation) | Adversarial patch attack against Jedi · Pascal VOC 07 | patch detection rate and recovery rate | Mixed |
| Use a GAN-based naturalistic patch | Jedi on detection benchmarks · CASIA and INRIA | recovery rate: up to 76% | Improved |
Later papers that compared with this method
5 later accepted papers report Jedi as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Real-World Adversarial Defense Against Patch Attacks Based on Diffusion Model (TPAMI 2025 · better on 5 of 5 shared results)
- PBCAT: Patch-Based Composite Adversarial Training against Physically Realizable Attacks on Object Detection (ICCV 2025 · better on 3 of 3 shared results)
- ODDR: Outlier Detection & Dimension Reduction Based Defense Against Adversarial Patches (ICCV 2025 · better on 2 of 2 shared results)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · better on 3 of 3 shared results)
- DIFFender: Diffusion-Based Adversarial Defense against Patch Attacks (ECCV 2024 · better on 3 of 3 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
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.90)
- Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights (ICCV 2025 · similarity 0.89)
- IAP: Invisible Adversarial Patch Attack through Perceptibility-Aware Localization and Perturbation Optimization (ICCV 2025 · similarity 0.89)
- False Positives Matter: Multidimensional Localization Evaluation and Training-Free Explainable Adversarial Patch Defense (AAAI 2026 · similarity 0.89)
- 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)
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.
