Research map / Label-efficient and robust learning / Recognition and long-tail
Id-Free Person Similarity Learning
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 |
|---|---|---|---|
| Add unlabeled Kinetics video fine-tuning with instance-level appearance changes | PointID trained on COCO and CrowdHuman · CUHK-SYSU | mAP: 76.5% -> 82.5% | Helped |
| Add unlabeled Kinetics video fine-tuning with instance-level appearance changes | PointID trained on COCO and CrowdHuman · CUHK-SYSU | Top-1: 78.0% -> 84.6% | Helped |
| Pretrain with id-free similarity learning | PointID · CUHK-SYSU | mAP: 92.7% -> 94.8% | Helped |
| Pretrain with id-free similarity learning | PointID · PRW | mAP: 41.3% -> 47.4% | Helped |
| Train with the proposed contrastive loss instead of cross entropy (ablation) | PointID similarity learning · CUHK-SYSU | mAP: 54.4% -> 75.5% | Helped |
| Use all person boxes rather than sampled part data for contrastive training (ablation) | PointID similarity learning · CUHK-SYSU | mAP: 75.5% -> 76.5% | Helped |
| Add instance-level video appearance augmentation instead of image-level augmentation during video fine-tuning (ablation) | PointID · CUHK-SYSU | mAP: 73.2% -> 85.5% | Helped |
| Sample frames from opposite halves of a video rather than two random frames (ablation) | PointID video fine-tuning · CUHK-SYSU | mAP: 84.4% -> 85.5% | Helped |
| Use non-memory contrastive learning rather than memory-based learning for video fine-tuning (ablation) | PointID · CUHK-SYSU | mAP: 84.3% -> 85.5% | Helped |
| Fine-tune on video frames without instance-level pairing, using image-level augmentation instead (ablation) | PointID video fine-tuning · CUHK-SYSU | mAP: 78.9% -> 73.2% | Hurt |
Closest papers by meaning
- Distilled Person Re-Identification: Towards a More Scalable System (CVPR 2019 · similarity 0.89)
- Person Re-Identification in the Wild (CVPR 2017 · similarity 0.88)
- Joint Detection and Identification Feature Learning for Person Search (CVPR 2017 · similarity 0.87)
- Unsupervised Pre-Training for Person Re-Identification (CVPR 2021 · similarity 0.87)
- Diverse Knowledge Distillation for End-to-End Person Search (AAAI 2021 · similarity 0.87)
- Divide and Conquer: Hybrid Pre-training for Person Search (AAAI 2024 · similarity 0.86)
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.