Research map / Label-efficient and robust learning / Knowledge distillation
Diverse Knowledge Distillation for End-to-End Person Search
Read the paperCheck 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 |
|---|---|---|---|
| Replace the plain end-to-end model with diverse knowledge distillation and spatial-invariant augmentation | End-to-end person search · PRW | mAP: 41.91 -> 50.51 | Helped |
| Replace the plain end-to-end model with diverse knowledge distillation and spatial-invariant augmentation | End-to-end person search · CUHK-SYSU (gallery size 100) | mAP: 86.47 -> 93.09 | Helped |
| Add image-level spatial-invariant augmentation (ISA) to the external Re-ID teacher (ablation) | Two-step person search · PRW | mAP: 51.85 -> 54.16 | Helped |
| Use an ISA-trained teacher instead of the unaugmented teacher for probability KD (ablation) | End-to-end person search · PRW | mAP: 49.44 -> 49.78 | Helped |
| Add feature-level spatial-invariant augmentation (FSA) (ablation) | Ours-2 · PRW | mAP: 49.78 -> 50.15 | Helped |
| Add pair-wise relation-aware KD (ablation) | Ours-0 · PRW | mAP: 41.91 -> 46.00 | Helped |
| Add triplet-wise relation-aware KD (ablation) | Ours-0 · PRW | mAP: 41.91 -> 45.54 | Helped |
| Add both pair-wise and triplet-wise relation-aware KD (ablation) | Ours-0 · PRW | mAP: 41.91 -> 47.18 | Helped |
| Add probability-aware KD to pair-wise relation-aware KD (ablation) | Ours-4 · PRW | mAP: 46.00 -> 50.18 | Helped |
| Add probability-aware KD to triplet-wise relation-aware KD (ablation) | Ours-5 · PRW | mAP: 45.54 -> 50.46 | Helped |
| Combine probability-aware, pair-wise, and triplet-wise KD (ablation) | Ours-0 · PRW | mAP: 41.91 -> 50.51 | Helped |
| Remove random erasing from the external Re-ID teacher (ablation) | External Re-ID model used by Ours · PRW | mAP: 50.51 -> 50.19 | Hurt |
Closest papers by meaning
- Sequential End-to-end Network for Efficient Person Search (AAAI 2021 · similarity 0.91)
- Query-Guided End-To-End Person Search (CVPR 2019 · similarity 0.91)
- Divide and Conquer: Hybrid Pre-training for Person Search (AAAI 2024 · similarity 0.90)
- Hierarchical Online Instance Matching for Person Search (AAAI 2020 · similarity 0.89)
- End-to-End Thorough Body Perception for Person Search (AAAI 2020 · similarity 0.89)
- Cheb-GR: Rethinking K-nearest Neighbor Search in Re-ranking for Person Re-identification (CVPR 2025 · similarity 0.88)
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.