Research map / Adversarial robustness and security / Attacks on text and graphs
Towards Adversarially Robust Object Detection
Haichao Zhang, Jianyu Wang
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 |
|---|---|---|---|
| adversarial training with task-oriented multi-task domain (MTD) instead of none | SSD VGG16 · VOC07 test, RAP attack eps 8 | mAP: 6.6 -> 44.9 | Improved |
| adversarial training with MTD | SSD VGG16 · VOC07 test, DAG attack eps 8 | mAP: 0.3 -> 28.5 | Improved |
| adversarial training with MTD | SSD VGG16 · VOC07 test, clean | mAP: 72.1 -> 48.0 | Hurt |
| use task-oriented domain MTD instead of task-agnostic domain CON (ablation) | SSD VGG16 · VOC07 test, loss_cls PGD attack | mAP: 18.3 -> 29.1 | Improved |
| use task-oriented domain MTD instead of task-agnostic domain CON (ablation) | SSD VGG16 · VOC07 test, loss_loc PGD attack | mAP: 27.2 -> 31.9 | Improved |
| use task-oriented domain MTD instead of task-agnostic domain CON (ablation) | SSD VGG16 · VOC07 test, clean | mAP: 38.7 -> 48.0 | Improved |
| use multi-task domain MTD instead of classification-only domain CLS (ablation) | SSD VGG16 · VOC07 test, loss_cls PGD attack | mAP: 21.8 -> 29.1 | Improved |
| use multi-task domain MTD instead of localization-only domain LOC (ablation) | SSD VGG16 · VOC07 test, DAG attack | mAP: 17.2 -> 28.5 | Improved |
| use multi-task domain MTD instead of CLS (ablation) | SSD VGG16 · VOC07 test, loss_loc PGD attack | mAP: 32.2 -> 31.9 | No clear effect |
| use CLS domain instead of LOC domain (ablation) | SSD VGG16 · VOC07 test, clean | mAP: 51.9 vs 46.7 | Mixed |
| adversarial training with MTD on ResNet50 backbone | SSD ResNet50 · VOC07 test, DAG attack | mAP: 0.4 -> 22.9 | Improved |
| adversarial training with MTD on DarkNet53 backbone | SSD DarkNet53 · VOC07 test, RAP attack | mAP: 8.2 -> 46.6 | Improved |
Closest papers by meaning
- Class-Aware Robust Adversarial Training for Object Detection (CVPR 2021 · similarity 0.92)
- Adversarially-Aware Robust Object Detector (ECCV 2022 · similarity 0.92)
- Transferable Adversarial Attacks for Object Detection Using Object-Aware Significant Feature Distortion (AAAI 2024 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Can't Slow Me Down: Learning Robust and Hardware-Adaptive Object Detectors against Latency Attacks for Edge Devices (CVPR 2025 · similarity 0.86)
- KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models (AAAI 2025 · 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.
