Research map / Label-efficient and robust learning / Recognition and long-tail
Hierarchical Gaussian Descriptor for Person Re-Identification
Tetsu Matsukawa, Takahiro Okabe, Einoshin Suzuki, Yoichi Sato
Check 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 |
|---|---|---|---|
| use hierarchical Gaussian modeling of patch and region distributions instead of a global covariance descriptor (ablation) | GOG · VIPeR | rank-1 identification rate: 23.6% -> 42.3% | Improved |
| add mean information to covariance-based hierarchical region modeling (ablation) | Cov-of-Gauss · VIPeR | rank-1 identification rate: 41.0% -> 42.3% | Improved |
| use Gaussian rather than covariance modeling at the patch level (ablation) | hierarchical descriptor · VIPeR | rank-1 identification rate: 27.8% -> 41.0% | Improved |
| add tangent-space mapping at both hierarchy levels (ablation) | GOG · VIPeR | rank-1 identification rate: 7.6% -> 42.3% | Improved |
| remove tangent-space mapping at both hierarchy levels (ablation) | GOG · VIPeR | rank-1 identification rate: 42.3% -> 7.6% | Hurt |
| use mean removal before L2 normalization (ablation) | GOG · VIPeR | rank-1 identification rate: 22.9% -> 42.3% | Improved |
| add vertical pixel location to gradient and RGB features (ablation) | GOG pixel features · VIPeR | rank-1 identification rate: 37.7% -> 42.3% | Improved |
| fuse GOG descriptors extracted from four color spaces (ablation) | GOG pixel features · VIPeR | rank-1 identification rate: 42.3% -> 49.4% | Improved |
| use GOGFusion+XQDA instead of MetricEnsemble | person re-identification · VIPeR | rank-1 identification rate: 49.7 vs 45.9 | Improved |
| use GOGFusion+XQDA instead of MetricEnsemble | person re-identification · CUHK01 (M=1) | rank-1 identification rate: 57.8 vs 53.4 | Improved |
| use GOGFusion+XQDA instead of LOMO+XQDA | person re-identification · PRID450S | rank-1 identification rate: 68.4 vs 62.6 | Improved |
| use GOGFusion+XQDA instead of LOMO+XQDA | person re-identification · GRID | rank-1 identification rate: 24.7 vs 16.6 | Improved |
Closest papers by meaning
- Hierarchical Gaussian Descriptors with Application to Person Re-Identification (TPAMI 2019 · similarity 0.97)
- Diversity Regularized Spatiotemporal Attention for Video-Based Person Re-Identification (CVPR 2018 · similarity 0.84)
- Learning Instance-Level Spatial-Temporal Patterns for Person Re-Identification (ICCV 2021 · similarity 0.84)
- Spatio-Temporal Representation Factorization for Video-Based Person Re-Identification (ICCV 2021 · similarity 0.84)
- Exploring the Camera Bias of Person Re-identification (ICLR 2025 · similarity 0.83)
- Learning Commonality, Divergence and Variety for Unsupervised Visible-Infrared Person Re-identification (NeurIPS 2024 · similarity 0.83)
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.
