Research map / Image generation / Editing and style transfer
Pose-Normalized Image Generation for Person Re-identification
Xuelin Qian, Yanwei Fu, Tao Xiang, Wenxuan Wang, Jie Qiu, Yang Wu, Yu-Gang Jiang, Xiangyang Xue
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 |
|---|---|---|---|
| add pose-normalized features from PN-GAN images to ResNet-50-A | ResNet-50-A · Market-1501 (single-query) | Rank-1 / mAP: 87.26 -> 89.43 R-1, 69.32 -> 72.58 mAP | Improved |
| add pose-normalized features from PN-GAN images to ResNet-50-A | ResNet-50-A · Market-1501 (multi-query) | Rank-1 / mAP: 91.81 -> 92.93 R-1, 77.85 -> 80.19 mAP | Improved |
| add pose-normalized features from PN-GAN images to ResNet-50-A | ResNet-50-A · DukeMTMC-reID (supervised) | Rank-1 / mAP: 72.80 -> 73.58 R-1, 52.48 -> 53.20 mAP | Improved |
| add pose-normalized features from PN-GAN images to ResNet-50-A | ResNet-50-A · CUHK03 (supervised) | Rank-1: 76.83 -> 79.76 | Improved |
| add pose-normalized features from PN-GAN images to ResNet-50-A | ResNet-50-A · CUHK01 (supervised) | Rank-1: 64.56 -> 67.65 | Improved |
| add pose-normalized features, transfer from Market-1501 without fine-tuning | ResNet-50-A (TL) · DukeMTMC-reID (transfer) | Rank-1 / mAP: 27.872 -> 29.937 R-1, 13.942 -> 15.768 mAP | Improved |
| add pose-normalized features, transfer from Market-1501 without fine-tuning | ResNet-50-A (TL) · CUHK03 (transfer) | Rank-1: 16.50 -> 16.85 | No clear effect |
| add pose-normalized features, transfer from Market-1501 without fine-tuning | ResNet-50-A (TL) · CUHK01 (transfer) | Rank-1: 27.20 -> 27.58 | No clear effect |
| add original-image features (ResNet-50-A) to pose-normalized features (ResNet-50-B) (ablation) | ResNet-50-B · Market-1501 | Rank-1 / mAP: 63.75 -> 89.43 R-1, 41.29 -> 72.58 mAP | Improved |
| add pose-normalized features (ResNet-50-B) to original-image features (ablation) | ResNet-50-A · Market-1501 | Rank-1 / mAP: 87.26 -> 89.43 R-1, 69.32 -> 72.58 mAP | Improved |
| use eight canonical poses instead of one pose (ablation) | Ours (SL) · Market-1501 | mAP: 69.60 -> 72.58 | Improved |
| use eight canonical poses instead of one pose (ablation) | Ours (SL) · Market-1501 | Rank-1: 87.65 -> 89.40 | Improved |
Closest papers by meaning
- FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification (NeurIPS 2018 · similarity 0.92)
- Learning Disentangled Representation for Robust Person Re-identification (NeurIPS 2019 · similarity 0.91)
- Appearance and Pose-Conditioned Human Image Generation Using Deformable GANs (TPAMI 2019 · similarity 0.91)
- Recover and Identify: A Generative Dual Model for Cross-Resolution Person Re-Identification (ICCV 2019 · similarity 0.90)
- Learning Invariance From Generated Variance for Unsupervised Person Re-Identification (TPAMI 2022 · similarity 0.90)
- Progressive Pose Attention Transfer for Person Image Generation (CVPR 2019 · similarity 0.89)
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.
