Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Scale Deep Learning Architectures for Person Re-Identification
Xuelin Qian, Yanwei Fu, Yu-Gang Jiang, Tao Xiang, 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 saliency-based learning fusion layer (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 77.88 -> 79.01 | Improved |
| add classification subnet (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 76.21 -> 79.01 | Improved |
| add both fusion layer and classification subnet (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 74.21 -> 79.01 | Improved |
| use multi-scale stream layers instead of Inception A+B (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 72.11 -> 79.01 | Improved |
| use multi-scale stream layers instead of Inception B (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 67.31 -> 79.01 | Improved |
| use multi-scale stream layers instead of Inception A (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: 60.11 -> 79.01 | Improved |
| use gradual Reduction layer instead of max-pooling (ablation) | MuDeep · CUHK01 | Rank-1 accuracy: more than 10% absolute points lower | Improved |
| use squared feature difference instead of plain difference (ablation) | MuDeep verification subnet · CUHK01 | Rank-1 accuracy: 2.2% higher | Improved |
| train jointly with CUHK03 labelled and detected data | MuDeep · CUHK03-Detected | Rank-1 accuracy: 75.34 -> 75.64 | Improved |
| train jointly with CUHK03 labelled and detected data | MuDeep · CUHK03-Labelled | Rank-1 accuracy: 76.34 -> 76.87 | Improved |
| fuse models trained at original and lower resolution | MuDeep · CUHK01 / VIPeR / CUHK03 | Rank-1 accuracy: around 1-2% lower | Hurt |
| add max-pooling frame aggregation for video re-id (no RNN) | MuDeep · PRID-2011 | Rank-1 accuracy: 65 vs 70 (RCNvrid) | Mixed |
Closest papers by meaning
- Leader-Based Multi-Scale Attention Deep Architecture for Person Re-Identification (TPAMI 2019 · similarity 0.94)
- Deep Learning for Person Re-Identification: A Survey and Outlook (TPAMI 2021 · similarity 0.93)
- Learning Generalisable Omni-Scale Representations for Person Re-Identification (TPAMI 2021 · similarity 0.92)
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.92)
- Resource Aware Person Re-Identification Across Multiple Resolutions (CVPR 2018 · similarity 0.92)
- Omni-Scale Feature Learning for Person Re-Identification (ICCV 2019 · similarity 0.91)
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.
