Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Group-Shuffling Random Walk for Person Re-Identification
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 |
|---|---|---|---|
| Use end-to-end group-shuffling random walk (ablation) | ResNet-50 pairwise affinity CNN · Market-1501 | mAP: 76.4 -> 82.5 | Helped |
| Use end-to-end group-shuffling random walk (ablation) | ResNet-50 pairwise affinity CNN · CUHK03 | mAP: 88.9 -> 94.0 | Helped |
| Use end-to-end group-shuffling random walk (ablation) | ResNet-50 pairwise affinity CNN · DukeMTMC | mAP: 61.8 -> 66.4 | Helped |
| Add end-to-end random walk rather than use no random walk (ablation) | Grouped affinity model · Market-1501, CUHK03, and DukeMTMC | mAP: 2.9%, 1.3%, and 2.7% | Helped |
| Add 4-way feature grouping without random walk (ablation) | Baseline pairwise affinity CNN · Market-1501 and CUHK03 | mAP: 76.4 -> 77.7; 88.9 -> 91.6 | Helped |
| Add group shuffle to 4-group end-to-end random walk (ablation) | Grouped random-walk model · Market-1501, CUHK03, and DukeMTMC | mAP: 81.6 -> 82.5; 92.9 -> 94.0; 65.4 -> 66.4 | Helped |
| Increase feature groups from 2 to 4 with random walk and no shuffle (ablation) | Grouped random-walk model · Market-1501, CUHK03, and DukeMTMC | mAP: 81.4 -> 81.6; 91.4 -> 92.9; 65.2 -> 65.4 | Mixed |
| Add feature dropout to baseline (ablation) | ResNet-50 pairwise affinity CNN · Market-1501 and CUHK03 | mAP: 76.4 -> 77.6; 88.9 -> 89.1 | Helped |
| Combine feature grouping with dropout (ablation) | Baseline pairwise affinity CNN · Market-1501 and CUHK03 | mAP: 77.7 -> 78.1; 91.6 -> 91.3 | Mixed |
| Train the network with random walk rather than apply random walk only at testing (ablation) | Baseline pairwise affinity CNN · Market-1501 and CUHK03 | mAP: 79.2 -> 81.4; 90.2 -> 91.5 | Helped |
| Replace end-to-end random walk with k-reciprocal re-ranking (ablation) | Baseline learned features · Market-1501 and CUHK03 | mAP: 81.4 vs 78.5; 91.5 vs 89.9 | Hurt |
| Add feature grouping and group shuffle during training, then evaluate feature distances (ablation) | Trained ResNet-50 features · Market-1501 and DukeMTMC | mAP: 74.6 -> 76.9; 60.3 -> 62.1 | Helped |
Closest papers by meaning
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.87)
- Cheb-GR: Rethinking K-nearest Neighbor Search in Re-ranking for Person Re-identification (CVPR 2025 · similarity 0.86)
- Joint Detection and Identification Feature Learning for Person Search (CVPR 2017 · similarity 0.85)
- Group Consistent Similarity Learning via Deep CRF for Person Re-Identification (CVPR 2018 · similarity 0.85)
- Diverse Knowledge Distillation for End-to-End Person Search (AAAI 2021 · similarity 0.85)
- Sequential End-to-end Network for Efficient Person Search (AAAI 2021 · 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.