Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Pre-Training for Person Re-Identification
Check 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 LUPerson unsupervised pre-training instead of ImageNet supervised pre-training | MGN, Trip, and IDE · CUHK03, Market1501, DukeMTMC, MSMT17 | mAP: +4.2%, +3.5%, +2.7%, +2% respectively | Helped |
| Use LUPerson unsupervised pre-training with SpCL | SpCL USL · Market1501 and DukeMTMC | mAP: 3.8 mAP and 2.2 mAP | Helped |
| Use LUPerson unsupervised pre-training with SpCL | SpCL UDA · DukeMTMC to Market1501 and Market1501 to DukeMTMC | mAP: 2.8% and 1.2% | Helped |
| Remove grayscale augmentation (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 73.4 -> 73.2 | No clear effect |
| Remove Gaussian blurring (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 73.4 -> 73.3 | No clear effect |
| Remove color jitter (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 73.4 -> 74.0 | Helped |
| Add RandomErasing (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 73.4 -> 74.2 | Helped |
| Set RandomErasing maximum area to 0.6 (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 73.2 -> 74.7 | Helped |
| Set RandomErasing maximum area to 0.8 rather than 0.6 (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 74.7 -> 73.3 | Hurt |
| Set contrastive-loss temperature to 0.07 rather than 0.2 (ablation) | MoCo v2 pre-training · CUHK03 | mAP: 71.1 -> 74.7 | Helped |
| Increase LUPerson pre-training data from 12.5% to 100% (ablation) | MGN · CUHK03 | mAP: 69.2/69.0 -> 74.7/75.4 | Helped |
| Replace ImageNet supervised pre-training with LUPerson unsupervised pre-training | MGN · Market1501 small-scale 10% | mAP: 53.1/76.9 -> 64.6/85.5 | Helped |
Closest papers by meaning
- Large-Scale Pre-Training for Person Re-Identification With Noisy Labels (CVPR 2022 · similarity 0.91)
- Resource Aware Person Re-Identification Across Multiple Resolutions (CVPR 2018 · similarity 0.90)
- Camera-Aware Proxies for Unsupervised Person Re-Identification (AAAI 2021 · similarity 0.90)
- Unsupervised Person Re-Identification by Soft Multilabel Learning (CVPR 2019 · similarity 0.90)
- Unsupervised Cross-Dataset Transfer Learning for Person Re-Identification (CVPR 2016 · similarity 0.90)
- Joint Discriminative and Generative Learning for Person Re-Identification (CVPR 2019 · similarity 0.90)
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.