Research map / Adversarial robustness and security / Attacks on text and graphs
ODDR: Outlier Detection & Dimension Reduction Based Defense Against Adversarial Patches
Nandish Chattopadhyay, Amira Guesmi, Muhammad Abdullah Hanif, Bassem Ouni, Muhammad Shafique
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 |
|---|---|---|---|
| ODDR defense added against GoogleAP | ResNet-152 · ImageNet | robust accuracy (%): 39.3 -> 79.1 | Improved |
| ODDR added against LAVAN | ResNet-152 · ImageNet | robust accuracy (%): 10.1 -> 74.1 | Improved |
| ODDR added against DorPatch | ResNet-50 · ImageNet | robust accuracy (%): 4.5 -> 59.8 | Improved |
| ODDR added against SSAP | MIMdepth · KITTI | mean depth estimation error: 0.59 -> 0.11 | Improved |
| ODDR instead of LGS | ResNet-152 GAP · ImageNet | robust accuracy (%): 53.9 -> 79.1 | Improved |
| ODDR instead of Dimension Reduction only (no outlier detection) (ablation) | ResNet-152 GAP · ImageNet | robust accuracy (%): 66.2 -> 79.1 | Improved |
| Isolation Forest instead of DBSCAN for outlier detection (ablation) | ODDR · ImageNet | overlap of detected cluster with patch: up to 20% | Improved |
| Dimension Reduction instead of naive pixel replacement (lowest, highest, mean) (ablation) | ODDR neutralization · ImageNet | accuracy: up to 15% | Improved |
| Adaptive distribution-mapping attack against ODDR | ODDR ResNet-50 · ImageNet | accuracy (%): clean 78 -> 73, robust 76 | Mixed |
| Zeroth-order gradient adaptive attack, delta 0.001-0.2 | ODDR · ImageNet | adversarial accuracy: within 5-10% of clean accuracy | No clear effect |
| Low-dimensional patch (99% variability preserved) adaptive attack | ODDR · ImageNet | attack effectiveness: 99% variability | Hurt |
| ODDR instead of Jedi on naturalistic patch | YOLO · CASIA | robust average precision: 67% -> 87.5% | Improved |
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.
- Jedi — Jedi: Entropy-Based Localization and Removal of Adversarial Patches (CVPR 2023) · better on 2 of 2 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
- RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples (ICML 2024 · similarity 0.88)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.88)
- Adversarially Robust Anomaly Detection through Spurious Negative Pair Mitigation (ICLR 2025 · similarity 0.88)
- Contamination-Resilient Anomaly Detection via Adversarial Learning on Partially-Observed Normal and Anomalous Data (ICML 2024 · similarity 0.86)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.85)
- NAPGuard: Towards Detecting Naturalistic Adversarial Patches (CVPR 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.
