Research map / Adversarial robustness and security / Certified and empirical robustness
Debiased Dual-Invariant Defense for Adversarially Robust Person Re-Identification
Yuhang Zhou, Yanxiang Zhao, Zhongyun Hua, Zhipu Liu, Zhaoquan Gu, Qing Liao, Leo Yu Zhang
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 |
|---|---|---|---|
| add diffusion-based data balancing (ablation) | Metric AT · Market-1501 FNA 8/255-16 ablation | mAP: 28.38 -> 29.85 | Improved |
| add adversarial feature learning (ablation) | Metric AT · Market-1501 FNA 8/255-16 ablation | mAP: 28.38 -> 30.34 | Improved |
| add self-meta learning (ablation) | Metric AT · Market-1501 FNA 8/255-16 ablation | mAP: 28.38 -> 29.48 | Improved |
| add FNES (farthest negative extension softening) (ablation) | Metric AT · Market-1501 FNA 8/255-16 ablation | mAP: 28.38 -> 30.98 | Improved |
| combine all modules (ablation) | Metric AT · Market-1501 FNA 8/255-16 ablation | mAP: 28.38 -> 31.99 | Improved |
| combine all modules (ablation) | Metric AT · DukeMTMC FNA 8/255-16 ablation | mAP: 23.76 -> 26.02 | Improved |
| add diffusion, adversarial and self-meta modules individually (ablation) | Metric AT · DukeMTMC white-box SMA 8/255-16 | mAP: 40.40 -> 39.44, 38.67, 38.65 | Hurt |
| use metric adversarial training instead of PGD classification AT (ablation) | AT_PGD · Market-1501 FNA 8/255-16 | mAP: 26.18 -> 28.38 | Improved |
| full method instead of Metric AT on unseen domain (ablation) | Metric AT · Market-1501 to DukeMTMC cross-domain FNA 8/255-16 | mAP: 4.47 -> 6.17 | Improved |
| fine-tune only classifier on clean data after adversarial training | ResNet18 AT · CIFAR10 PGD | robust accuracy: 54.045 -> 50.752 | Hurt |
| fine-tune only classifier on clean data after adversarial training | ResNet50 AT · CIFAR10 PGD | robust accuracy: 53.076 -> 50.218 | Hurt |
| apply the full defense instead of standard adversarial training | Adv_train · Market-1501 clean | mAP: 69.69 -> 68.50 | Hurt |
Closest papers by meaning
- Transferable, Controllable, and Inconspicuous Adversarial Attacks on Person Re-identification With Deep Mis-Ranking (CVPR 2020 · similarity 0.90)
- Towards Robust Person Re-Identification by Defending Against Universal Attackers (TPAMI 2022 · similarity 0.90)
- Exploring the Camera Bias of Person Re-identification (ICLR 2025 · similarity 0.88)
- Feature-Level Adversarial Attacks and Ranking Disruption for Visible-Infrared Person Re-identification (NeurIPS 2024 · similarity 0.87)
- From Poses to Identity: Training-Free Person Re-Identification via Feature Centralization (CVPR 2025 · similarity 0.86)
- Person De-reidentification: A Variation-guided Identity Shift Modeling (CVPR 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.
