Research map / Label-efficient and robust learning / Recognition and long-tail
Learning a Key-Value Memory Co-Attention Matching Network for Person Re-Identification
Yaqing Zhang, Xi Li, Zhongfei 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 |
|---|---|---|---|
| use 2 heads instead of 1 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 85.3 -> 86.9 | Improved |
| use 4 heads instead of 1 (base) (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 85.3 -> 87.1 | Improved |
| increase heads from 4 to 8 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 86.2 | Hurt |
| replace key-value with K = V (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 83.5 | Hurt |
| increase feature dimension d from 256 to 1024 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 88.3 | Improved |
| remove position embedding (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 86.8 | No clear effect |
| training gallery set M=2 random negatives instead of 10 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 68.4 | Hurt |
| use hard negatives for M=2 instead of random (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 68.4 -> 78.0 | Improved |
| increase gallery size M from 10 to 15 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 88.0 | Improved |
| increase gallery size M from 15 to 30 (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 88.0 -> 87.6 | No clear effect |
| pairwise metric learning instead of setwise matching network (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 88.0 vs 85.4 | Hurt |
| add random erasing augmentation (ablation) | KVM-MN · CUHK03 labeled (1 split, ablation) | rank-1: 87.1 -> 88.2 | Improved |
Closest papers by meaning
- Harmonious Attention Network for Person Re-Identification (CVPR 2018 · similarity 0.89)
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.88)
- Dual Attention Matching Network for Context-Aware Feature Sequence Based Person Re-Identification (CVPR 2018 · similarity 0.88)
- Re-Identification With Consistent Attentive Siamese Networks (CVPR 2019 · similarity 0.88)
- Multi-Channel Pyramid Person Matching Network for Person Re-Identification (AAAI 2018 · similarity 0.88)
- Consistent-Aware Deep Learning for Person Re-Identification in a Camera Network (CVPR 2017 · similarity 0.88)
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.
