Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Generalisable Omni-Scale Representations for Person Re-Identification
Kaiyang Zhou, Yongxin Yang, Andrea Cavallaro, Tao Xiang
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 omni-scale streams T=4 instead of single-scale T=1 (ablation) | OSNet · MSMT17 | Rank-1: 52.9 -> 71.0 | Improved |
| use unified aggregation gate instead of addition (ablation) | OSNet · MSMT17 | Rank-1: 64.5 -> 71.0 | Improved |
| use unified aggregation gate instead of concatenation (ablation) | OSNet · MSMT17 | Rank-1: 66.4 -> 71.0 | Improved |
| share one aggregation gate across streams instead of separate AGs (ablation) | OSNet · MSMT17 | Rank-1: 68.3 -> 71.0 | Improved |
| use channel-wise gates instead of stream-wise gates (ablation) | OSNet · MSMT17 | Rank-1: 67.9 -> 71.0 | Improved |
| use dynamic input-dependent gates instead of learned-and-fixed gates (ablation) | OSNet · MSMT17 | Rank-1: 64.5 -> 71.0 | Improved |
| use heterogeneous-depth streams instead of same-depth ResNeXt-like streams (ablation) | OSNet · CUHK03 | Rank-1: 54.4 -> 57.1 | Improved |
| use Lite 3x3 factorised convolutions instead of standard convolutions (ablation) | OSNet · MSMT17 | Rank-1: 70.5 -> 71.0 | No clear effect |
| add instance normalisation via architecture search (OSNet-AIN vs OSNet) | OSNet · Market1501->Duke | Rank-1: 44.7 -> 52.4 | Improved |
| add instance normalisation via architecture search (OSNet-AIN vs OSNet) | OSNet · Duke->Market1501 | Rank-1: 52.2 -> 61.0 | Improved |
| search IN placement instead of hand-placing IN in lowest layers (ablation) | OSNet-IN · Market1501->Duke | Rank-1: 47.9 -> 52.4 | Improved |
| search IN placement instead of hand-placing IN in lowest layers (ablation) | OSNet-IN · Duke->Market1501 | Rank-1: 57.8 -> 61.0 | Improved |
Closest papers by meaning
- Omni-Scale Feature Learning for Person Re-Identification (ICCV 2019 · similarity 0.97)
- Multi-Scale Deep Learning Architectures for Person Re-Identification (ICCV 2017 · similarity 0.92)
- Leader-Based Multi-Scale Attention Deep Architecture for Person Re-Identification (TPAMI 2019 · similarity 0.92)
- Deep Learning for Person Re-Identification: A Survey and Outlook (TPAMI 2021 · similarity 0.90)
- Resource Aware Person Re-Identification Across Multiple Resolutions (CVPR 2018 · similarity 0.90)
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · 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.
