Research map / Label-efficient and robust learning / Recognition and long-tail
Online Pseudo Label Generation by Hierarchical Cluster Dynamics for Adaptive Person Re-Identification
Yi Zheng, Shixiang Tang, Guolong Teng, Yixiao Ge, Kaijian Liu, Jing Qin, Donglian Qi, Dapeng Chen
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 cluster split to merge-only hierarchical clustering (ablation) | Ours · Duke to Market / Market to Duke | mAP (%): 75.0 -> 80.0 (D->M); 61.7 -> 70.1 (M->D) | Improved |
| add label propagation to average-linkage hierarchical clustering (ablation) | Ours · Duke to Market / Market to Duke | mAP (%): 53.3 -> 80.0 (D->M); 59.0 -> 70.1 (M->D) | Improved |
| add hierarchical levels to single-level label propagation (ablation) | Ours · Duke to Market / Market to Duke | mAP (%): 74.6 -> 80.0 (D->M); 68.9 -> 70.1 (M->D) | Improved |
| treat each target instance as its own class instead of clustering (ablation) | Src. only baseline · Duke to Market | mAP (%): 19.6 -> 6.0 | Hurt |
| increase label update interval from online (0) to 3 (ablation) | Ours · Duke to Market / Market to Duke | mAP (%): 80.0 -> 64.2 (D->M); 70.1 -> 62.2 (M->D) | Hurt |
| select anchors by density peak instead of random choosing (ablation) | Ours · Duke to Market | mAP (%): 75.0 -> 80.0 | Improved |
| select anchors by density peak instead of farthest point sampling (ablation) | Ours · Duke to Market | mAP (%): 75.4 -> 80.0 | Improved |
| raise merge alpha from 0.1 to 0.99 (ablation) | Ours · Duke to Market | mAP (%): 66.9 -> 80.0 | Improved |
| raise split alpha from 0.3 to 0.99 (ablation) | Ours · Duke to Market | mAP (%): 76.1 -> 80.0 | Improved |
| increase number of hierarchy levels from 1 to 2 (ablation) | Ours · Duke to Market / Market to Duke | mAP (%) | Improved |
| increase number of hierarchy levels beyond 2 (ablation) | Ours · Duke to Market / Market to Duke | mAP (%) | No clear effect |
| use online hierarchical cluster dynamics instead of epoch-wise DBSCAN (SpCL), unsupervised | Ours · Market-1501 unsupervised | mAP (%): 73.1 -> 78.1 | Improved |
Closest papers by meaning
- Group-aware Label Transfer for Domain Adaptive Person Re-identification (CVPR 2021 · similarity 0.93)
- Exploiting Sample Uncertainty for Domain Adaptive Person Re-Identification (AAAI 2021 · similarity 0.91)
- Reliability Exploration with Self-Ensemble Learning for Domain Adaptive Person Re-identification (AAAI 2022 · similarity 0.91)
- Delving into Probabilistic Uncertainty for Unsupervised Domain Adaptive Person Re-identification (AAAI 2022 · similarity 0.90)
- Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification (CVPR 2020 · similarity 0.89)
- Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification (ICLR 2020 · similarity 0.89)
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.
