Research map / Label-efficient and robust learning / Recognition and long-tail
Discrepant and Multi-Instance Proxies for Unsupervised Person Re-Identification
Chang Zou, Zeqi Chen, Zhichao Cui, Yuehu Liu, Chi Zhang
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 uni-proxy with two discrepant cluster proxies (ablation) | Cluster-Contrast style baseline · Market-1501 | mAP (%): 81.0 -> 85.8 | Improved |
| replace uni-proxy with two discrepant cluster proxies (ablation) | baseline · MSMT17 | mAP (%): 35.2 -> 38.4 | Improved |
| add multi-instance proxies with global hard negative mining (ablation) | DCMIP · MSMT17 | mAP (%): 37.5 -> 40.9 | Improved |
| use global hard negatives from multi-instance proxies instead of batch hard triplet mining (ablation) | DCMIP · MSMT17 | mAP (%): 37.9 -> 40.9 | Improved |
| use global hard negatives instead of batch hard triplet mining (ablation) | DCMIP · Market-1501 | mAP (%): 86.1 -> 86.7 | Improved |
| use three cluster proxies instead of two (ablation) | DCP · MSMT17 | mAP (%): 38.4 -> 37.5 | Hurt |
| use Hard update alone for the single cluster proxy (ablation) | uni-proxy · MSMT17 | mAP (%): 35.2 -> 18.5 | Hurt |
| use Mean+Hard proxies rather than Mean+Rand (ablation) | DCP · Market-1501 vs MSMT17 | mAP (%): Market-1501 83.1 -> 85.8; MSMT17 38.4 -> 33.9 | Mixed |
| increase number of global hard negative instance proxies N beyond 256 (ablation) | DCMIP · Market-1501 and MSMT17 | mAP (%): Market-1501 86.7 -> 86.4 (N=1024); MSMT17 40.9 -> 39.5 (N=1024) | Hurt |
| use DCMIP instead of Cluster-Contrast uni-proxy | unsupervised re-ID · MSMT17 | mAP (%): 33.0 -> 40.9 | Improved |
Closest papers by meaning
- Camera-Aware Proxies for Unsupervised Person Re-Identification (AAAI 2021 · similarity 0.91)
- Towards Discriminative Representation Learning for Unsupervised Person Re-Identification (ICCV 2021 · similarity 0.90)
- ICE: Inter-Instance Contrastive Encoding for Unsupervised Person Re-Identification (ICCV 2021 · similarity 0.89)
- Self-Similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-Identification (ICCV 2019 · similarity 0.89)
- Unsupervised Person Re-Identification via Softened Similarity Learning (CVPR 2020 · similarity 0.88)
- Learning Commonality, Divergence and Variety for Unsupervised Visible-Infrared Person Re-identification (NeurIPS 2024 · similarity 0.87)
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.
