Research map / Label-efficient and robust learning / Domain adaptation and generalization
Generalizable Person Re-Identification With Relevance-Aware Mixture of Experts
Yongxing Dai, Xiaotong Li, Jun Liu, Zekun Tong, Ling-Yu Duan
Check 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 |
|---|---|---|---|
| Use relevance-aware expert voting instead of the hybrid-dataset baseline | RaMoE · PRID, GRID, VIPeR, and iLIDs, Protocol-2 | mAP: 58.9 -> 66.8; 47.7 -> 53.9; 63.8 -> 72.2; 89.2 -> 92.3 | Improved |
| Use RaMoE rather than the baseline | RaMoE · Market, Duke, CUHK03, and MSMT17, Protocol-3 | mAP: 49.9 -> 56.5; 49.4 -> 56.9; 32.6 -> 35.5; 9.9 -> 13.5 | Improved |
| Add relevance-aware voting over expert features rather than direct concatenation (ablation) | RaMoE · PRID, GRID, VIPeR, and iLIDs, Protocol-2 | mAP: 65.1 -> 66.8; 52.3 -> 53.9; 70.6 -> 72.2; 91.4 -> 92.3 | Improved |
| Add the decorrelation loss to diversify source experts (ablation) | RaMoE · PRID, GRID, VIPeR, and iLIDs, Protocol-2 | mAP: 66.0 -> 66.8; 53.2 -> 53.9; 71.3 -> 72.2; 91.2 -> 92.3 | Improved |
| Use concatenation rather than element-wise summation for expert features (ablation) | RaMoE · GRID, Protocol-2 | mAP: 52.3 -> 53.9 (softmax); 51.9 -> 53.7 (sigmoid) | Improved |
| Replace softmax with sigmoid for relevance weights while concatenating features (ablation) | RaMoE · GRID, Protocol-2 | mAP: 53.9 -> 53.7 | No clear effect |
| Calculate test-domain relevance using gallery samples only (ablation) | RaMoE · Market, Duke, CUHK03, and MSMT17, Protocol-3 | mAP: 56.5 -> 56.3; 56.9 -> 56.8; 35.5 -> 35.4; 13.5 -> 13.5 | No clear effect |
| Use the combined experts rather than any single source-domain expert (ablation) | RaMoE · PRID, Protocol-2 | mAP: 62.2 -> 65.1; 61.6 -> 65.1; 62.5 -> 65.1; 63.6 -> 65.1 | Improved |
Closest papers by meaning
- Adaptive Cross-Domain Learning for Generalizable Person Re-identification (ECCV 2022 · similarity 0.92)
- Mimic Embedding via Adaptive Aggregation: Learning Generalizable Person Re-identification (ECCV 2022 · similarity 0.92)
- Meta Distribution Alignment for Generalizable Person Re-Identification (CVPR 2022 · similarity 0.90)
- Generalizable Person Re-Identification by Domain-Invariant Mapping Network (CVPR 2019 · similarity 0.90)
- Style Normalization and Restitution for Generalizable Person Re-Identification (CVPR 2020 · similarity 0.90)
- Generalizable Person Re-identification via Balancing Alignment and Uniformity (NeurIPS 2024 · 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.
