Research map / Label-efficient and robust learning / Recognition and long-tail
Hard-Aware Point-to-Set Deep Metric for Person Re-identification
Rui Yu, Zhiyong Dou, Song Bai, Zhaoxiang Zhang, Yongchao Xu, Xiang Bai
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 |
|---|---|---|---|
| replace hardest-sample selection with exponential soft hard-aware weighting (sigma=0.5) | Hard Triplet · Market-1501 | mAP: 67.22 -> 69.76 | Improved |
| replace hardest-sample selection with polynomial soft hard-aware weighting (alpha=10) | Hard Triplet · Market-1501 | mAP: 67.22 -> 69.43 | Improved |
| use uniform weights over the set (sigma -> infinity) instead of hard-aware weights (ablation) | HAP2S E · Market-1501 | mAP: 69.76 -> 1.36 | Hurt |
| use uniform weights (alpha=0) instead of hard-aware polynomial weights (ablation) | HAP2S P · Market-1501 | mAP: 69.43 -> 3.11 | Hurt |
| increase sigma from 0.5 to 8 (ablation) | HAP2S E · Market-1501 | mAP: 69.76 -> 62.30 | Hurt |
| lower alpha from 10 to 4 (ablation) | HAP2S P · Market-1501 | mAP: 69.43 -> 59.07 | Hurt |
| swap exponential for polynomial weighting (ablation) | HAP2S · Market-1501 | rank-1: 84.20 -> 84.59 | No clear effect |
| add 1000 false-positive outliers to the training set (ablation) | HAP2S E · Market-1501 | mAP: 64.48 | Hurt |
| add more than 200 outliers to the training set (ablation) | Hard Triplet · Market-1501 | training stability | Hurt |
| add re-ranking at test time | HAP2S · Market-1501 (single query) | mAP / rank-1: 79.91% mAP and 85.72% rank-1 | Improved |
| use HAP2S loss instead of Hard Triplet loss | ResNet-50 · CUHK03 (labeled) | rank-1: 87.65 -> 90.22 | Improved |
| use HAP2S E instead of Angular Loss on generic metric learning | ResNet-50 · CUB-200-2011 | Recall@1: 54.70 -> 56.08 | Improved |
Closest papers by meaning
- Deep Learning for Person Re-Identification: A Survey and Outlook (TPAMI 2021 · similarity 0.91)
- Point to Set Similarity Based Deep Feature Learning for Person Re-Identification (CVPR 2017 · similarity 0.90)
- Embedding Deep Metric for Person Re-identification: A Study Against Large Variations (ECCV 2016 · similarity 0.90)
- A Systematic Evaluation and Benchmark for Person Re-Identification: Features, Metrics, and Datasets (TPAMI 2018 · similarity 0.89)
- Unsupervised Person Re-Identification by Deep Asymmetric Metric Embedding (TPAMI 2018 · similarity 0.89)
- Resource Aware Person Re-Identification Across Multiple Resolutions (CVPR 2018 · 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.
