Research map / Video / Video segmentation and tracking
HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReID
Yiyang Su, Yunping Shi, Feng Liu, Xiaoming Liu
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 |
|---|---|---|---|
| replace vanilla MoE with hierarchical biometric experts (no consistency losses) (ablation) | HAMoBE · MEVID same clothes | top-1 (%): 63.0 -> 68.1 | Improved |
| replace vanilla MoE with hierarchical biometric experts (no consistency losses) (ablation) | HAMoBE · MEVID different clothes | top-1 (%): 14.3 -> 13.5 | Hurt |
| add consistency losses and contrastive loss (ablation) | HAMoBE · MEVID same clothes | top-1 (%): 68.1 -> 74.0 | Improved |
| add consistency losses and contrastive loss (ablation) | HAMoBE · MEVID different clothes | top-1 (%): 13.5 -> 20.1 | Improved |
| add dual-input decision mechanism to single-input mode (ablation) | HAMoBE · CCVID different clothes | top-1 (%): 85.7 -> 87.6 | Improved |
| combine long-term, short-term and temporal experts instead of temporal expert alone (ablation) | HAMoBE · CCVID different clothes | top-1 (%): 85.4 -> 87.6 | Improved |
| increase first-layer experts from 4 to 8 (ablation) | HAMoBE · CCVID different clothes | top-1 (%): 82.6 -> 85.8 | Improved |
| increase first-layer experts from 8 to 16 (ablation) | HAMoBE · CCVID different clothes | top-1 (%): 85.8 -> 84.5 | Hurt |
| use Transformer decoder instead of Transformer encoder for temporal aggregation (ablation) | HAMoBE temporal expert · MEVID different clothes | top-1 (%): 15.9 -> 20.1 | Improved |
| use Transformer decoder instead of mean pooling for temporal aggregation (ablation) | HAMoBE temporal expert · MEVID different clothes | top-1 (%): 9.8 -> 20.1 | Improved |
| use HAMoBE instead of TF-CLIP | HAMoBE · LS-VID | mAP (%): 83.8 -> 85.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TF-CLIP — TF-CLIP: Learning Text-Free CLIP for Video-Based Person Re-identification (AAAI 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Person Re-Identification by Discriminative Selection in Video Ranking (TPAMI 2016 · similarity 0.87)
- Instruct-ReID: A Multi-purpose Person Re-identification Task with Instructions (CVPR 2024 · similarity 0.86)
- Distilling CLIP with Dual Guidance for Learning Discriminative Human Body Shape Representation (CVPR 2024 · similarity 0.84)
- From Poses to Identity: Training-Free Person Re-Identification via Feature Centralization (CVPR 2025 · similarity 0.84)
- Exploring the Camera Bias of Person Re-identification (ICLR 2025 · similarity 0.84)
- CLIMB-ReID: A Hybrid CLIP-Mamba Framework for Person Re-Identification (AAAI 2025 · similarity 0.84)
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.
