Research map / Label-efficient and robust learning / Recognition and long-tail
Deeply-Learned Part-Aligned Representations for Person Re-Identification
Liming Zhao, Xi Li, Yueting Zhuang, Jingdong Wang
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 learned body part maps instead of 5 horizontal stripes (ablation) | Part-aligned representation network · Market-1501 | rank-1 (%): 74.1 -> 81.0 | Improved |
| use learned body part maps instead of 5x5 grid (ablation) | Part-aligned representation network · Market-1501 | rank-1 (%): 73.4 -> 81.0 | Improved |
| use learned body part maps instead of 5x5 grid (ablation) | Part-aligned representation network · CUHK03 (labeled) | rank-1 (%): 78.2 -> 85.4 | Improved |
| add the part net instead of a fully connected layer (ablation) | Part-aligned representation network · Market-1501 | mAP (%): 54.3 -> 63.4 | Improved |
| add the part net instead of global average pooling (ablation) | Part-aligned representation network · Market-1501 | mAP (%): 55.6 -> 63.4 | Improved |
| use 6 jointly learned parts instead of masks from a separate part segmentation model (ablation) | Part-aligned representation network · Market-1501 (6 parts) | rank-1 (%): 61.2 -> 80.4 | Improved |
| use 6 jointly learned parts instead of masks from a separate part segmentation model (ablation) | Part-aligned representation network · CUHK03 (6 parts) | rank-1 (%): 70.7 -> 85.1 | Improved |
| use 8 body parts instead of a single part (ablation) | Part-aligned representation network · CUHK03 (validation split) | rank-1 (%): 77.7 -> 83.8 | Improved |
| increase parts from 8 to 12 (ablation) | Part-aligned representation network · CUHK03 (validation split) | rank-1 (%): 83.8 -> 83.6 | No clear effect |
| use multiple body parts instead of human segmentation with 1 part (ablation) | Part-aligned representation network · Market-1501 | mAP (%): 58.9 -> 63.4 | Improved |
| use sigmoid instead of softmax for spatial part maps (ablation) | Part-aligned representation network · Market-1501 (AlexNet) | rank-1 (%): 62.23 -> 64.22 | Improved |
| remove pool3 to double the feature map size (ablation) | Part-aligned representation network · CUHK01 (486 test IDs) | rank-1 (%): 72.3 -> 75.0 | Improved |
Closest papers by meaning
- Part-Aligned Bilinear Representations for Person Re-Identification (ECCV 2018 · similarity 0.92)
- Pose-Driven Deep Convolutional Model for Person Re-Identification (ICCV 2017 · similarity 0.92)
- Multi-Task Learning With Coarse Priors for Robust Part-Aware Person Re-Identification (TPAMI 2020 · similarity 0.91)
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.91)
- Beyond Human Parts: Dual Part-Aligned Representations for Person Re-Identification (ICCV 2019 · similarity 0.91)
- Learning Deep Context-Aware Features Over Body and Latent Parts for Person Re-Identification (CVPR 2017 · 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.
