Research map / Adversarial robustness and security / Attacks on text and graphs
Class-Aware Robust Adversarial Training for Object 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 class-wise rather than object-wise adversarial training | SSD VGG16-BN · MS-COCO 2017 validation | mAP under PGD-10 A_cls: 0.100 -> 0.142 | Helped |
| use class-wise rather than object-wise adversarial training | SSD VGG16-BN · MS-COCO 2017 validation | mAP under CWA: 0.074 -> 0.092 | Helped |
| use class-wise rather than object-wise adversarial training | SSD VGG16-BN · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.229 -> 0.224 | Hurt |
| use class-wise rather than object-wise adversarial training | SSD VGG16-BN · PASCAL VOC 2007 test | mAP under CWA: 0.203 -> 0.199 | Hurt |
| use class-wise rather than object-wise adversarial training | SSD VGG16-BN · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.229 vs 0.224 on VOC; 0.100 vs 0.142 on COCO | Mixed |
| enable fast adversarial training | CWAT · training speed | speedup over MTD: 3.19x | Helped |
| clip task losses rather than leave them unbounded (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.173 -> 0.218 | Helped |
| set both clipping thresholds to 2 rather than 6 (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.218 -> 0.191 | Hurt |
| set both clipping thresholds to 4 rather than 6 (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.218 -> 0.190 | Hurt |
| set both clipping thresholds to 8 rather than 6 (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.218 -> 0.204 | Hurt |
| set classification and regression clipping thresholds to 2 and 6 rather than both 6 (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_cls: 0.218 -> 0.226 | Helped |
| use unequal clipping thresholds for classification and regression losses (ablation) | TOAT on PASCAL VOC 2007 test · PASCAL VOC 2007 test | mAP under PGD-10 A_reg: TOAT-2: 0.297; TOAT-4: 0.327; TOAT-2-4: 0.292; TOAT-4-2: 0.337 | Mixed |
Closest papers by meaning
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Transferable Adversarial Attacks for Object Detection Using Object-Aware Significant Feature Distortion (AAAI 2024 · similarity 0.87)
- KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models (AAAI 2025 · similarity 0.86)
- Identifying and Understanding Cross-Class Features in Adversarial Training (ICML 2025 · similarity 0.85)
- Adversarially Robust Few-shot Learning via Parameter Co-distillation of Similarity and Class Concept Learners (CVPR 2024 · similarity 0.85)
- Unsupervised Object Detection with Theoretical Guarantees (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.