Research map / Label-efficient and robust learning / Recognition and long-tail
Leveraging Prior Knowledge of Diffusion Model for Person Search
Giyeol Kim, Sooyoung Yang, Jihyong Oh, Myungjoo Kang, Chanho Eom
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 MSFRN to the re-ID branch (ablation) | DiffPS · PRW | mAP: 59.1 -> 61.6 | Improved |
| add SFAN to the baseline re-ID branch (ablation) | DiffPS · PRW | mAP: 59.1 -> 59.6 | Improved |
| add SFAN on top of MSFRN (ablation) | DiffPS · PRW | mAP: 61.6 -> 62.0 | Improved |
| add DGRPN to Faster R-CNN (ablation) | DiffPS detection branch · PRW | AP: 94.2 -> 94.8 | Improved |
| raise DGRPN threshold tau from 0.5 to 0.8 (ablation) | DGRPN · PRW | recall: 98.1 -> 97.6 | Hurt |
| lower DGRPN threshold tau from 0.5 to 0.2 (ablation) | DGRPN · PRW | AP: 94.8 -> 94.4 | Hurt |
| combine levels 1+2+3 features instead of level 2 only for re-ID (ablation) | DiffPS re-ID branch · PRW | mAP: 58.0 -> 62.0 | Improved |
| use all levels combined instead of levels 1+2+3 (ablation) | DiffPS re-ID branch · PRW | mAP: 62.0 -> 60.5 | Hurt |
| use SD v2-1 backbone instead of DINO ViT-G (ablation) | DiffPS · PRW | mAP: 41.5 -> 62.0 | Improved |
| use clothing prompts shirts/pants/shoes in SFAN instead of generic body parts (ablation) | SFAN · PRW | mAP: 61.5 -> 62.0 | Improved |
| add MSFRN to ResNet50 and SD v2-1 to reduce shape bias (ablation) | MSFRN · Cue-conflict | shape accuracy: 63.52 -> 58.28 | Improved |
| decouple detection and re-ID via frozen diffusion backbone instead of joint DMRNet | DiffPS · CUHK-SYSU | detection AP: 86.6 -> 90.9 | Improved |
Closest papers by meaning
- Diverse Knowledge Distillation for End-to-End Person Search (AAAI 2021 · similarity 0.89)
- OIMNet++: Prototypical Normalization and Localization-Aware Learning for Person Search (ECCV 2022 · similarity 0.88)
- Divide and Conquer: Hybrid Pre-training for Person Search (AAAI 2024 · similarity 0.87)
- Query-Guided End-To-End Person Search (CVPR 2019 · similarity 0.87)
- Joint Detection and Identification Feature Learning for Person Search (CVPR 2017 · similarity 0.87)
- Cheb-GR: Rethinking K-nearest Neighbor Search in Re-ranking for Person Re-identification (CVPR 2025 · 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.
