Research map / Label-efficient and robust learning / Recognition and long-tail
Discriminative Feature Learning With Consistent Attention Regularization for Person Re-Identification
Sanping Zhou, Fei Wang, Zeyi Huang, Jinjun 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 |
|---|---|---|---|
| add basic triplet loss to softmax loss (multi-task) (ablation) | ResNet50 softmax · Market1501 single-query | Top-1: 87.5 -> 89.1 | Improved |
| replace basic triplet loss with improved triplet loss (ablation) | ResNet50 + attention module + BT · Market1501 single-query | Top-1 / mAP: 3.1% in Top 1 and 4.1% in mAP | Improved |
| add attention module (without consistent attention regularizer) (ablation) | ResNet50 variants · Market1501 single-query | Top-1: 0.3%, 0.1%, 0.3%, 0.5% and 0.5% in Top 1 | No clear effect |
| add consistent attention regularizer to attention network (ablation) | ResNet50 + AM + S + IT · Market1501 single-query | Top-1: 93.9 -> 96.1 | Improved |
| add consistent attention regularizer to attention network (ablation) | ResNet50 + AM + S · Market1501 single-query | Top-1: 87.8 -> 89.3 | Improved |
| add consistent attention regularizer to attention network (ablation) | ResNet50 + AM + S + IT · Market1501 single-query | mAP: 79.5 -> 84.7 | Improved |
| add consistent attention regularizer to attention network (ablation) | ResNet50 + AM + S + IT · CUHK03 detected | Top-1: 85.5 -> 93.2 | Improved |
| replace basic triplet loss with improved triplet loss in full model (ablation) | full model · Market1501 single-query | Top-1: 92.1 -> 96.1 | Improved |
| replace symmetric triplet loss with improved triplet loss in full model (ablation) | full model · Market1501 single-query | Top-1: 94.2 -> 96.1 | Improved |
| use a large sparsity weight in the regularizer (ablation) | full model · Market1501 single-query | Top-1 / mAP | Hurt |
| vary alpha, beta and margin m (ablation) | full model · Market1501 single-query | Top-1 / mAP | No clear effect |
| use full method instead of MGN | MGN · Market1501 single-query | mAP: 86.9 -> 84.7 | Hurt |
Closest papers by meaning
- Scalable Person Re-Identification by Harmonious Attention (IJCV 2019 · similarity 0.91)
- Leader-Based Multi-Scale Attention Deep Architecture for Person Re-Identification (TPAMI 2019 · similarity 0.90)
- Harmonious Attention Network for Person Re-Identification (CVPR 2018 · similarity 0.90)
- Learning Generalisable Omni-Scale Representations for Person Re-Identification (TPAMI 2021 · similarity 0.90)
- Backbone Cannot Be Trained at Once: Rolling Back to Pre-Trained Network for Person Re-Identification (AAAI 2019 · similarity 0.90)
- Self-Training With Progressive Augmentation for Unsupervised Cross-Domain Person Re-Identification (ICCV 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.
