Research map / Label-efficient and robust learning / Recognition and long-tail
Spherical Feature Transform for Deep Metric Learning
Yuke Zhu, Yan Bai, Yichen Wei
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 SFT | NPair loss (GoogleNet) · Cars196 | Recall@1: 72.8 -> 79.4 | Improved |
| add SFT | Triplet loss (GoogleNet) · CUB-200-2011 | Recall@1: 42.8 -> 48.3 | Improved |
| add SFT | Multi-Similarity loss (GoogleNet-V2) · Cars196 | Recall@1: 84.0 -> 84.5 | Improved |
| add SFT | Ranked List Loss (GoogleNet-V2) · Cars196 | Recall@1: 74.2 -> 80.2 | Improved |
| add SFT | Triplet, RLL, MS · SOP | Recall@1: 70.8 -> 72.3; 77.5 -> 77.8; 73.1 -> 73.4 | No clear effect |
| use full SFT instead of degenerated SFT-d (ablation) | NPair loss (GoogleNet) · Cars196 | Recall@1: 76.2 -> 79.4 | Improved |
| use SFT instead of FTL feature transform | Triplet loss (GoogleNet) · Cars196 | Recall@1: 60.1 -> 65.1 | Improved |
| add SFT | CosFace on MS-Celeb-1M subset · YTF | verification accuracy (%): 96.2 -> 97.2 | Improved |
| add SFT | CosFace · IJB-C TAR@FAR 0.001% | TAR (%): 85.67 -> 87.19 | Improved |
| add SFT with unbalanced train scheme (ablation) | Baseline face network · Long-tailed MS-Celeb-1M subset (tail classes) | accuracy (%): 73.91 -> 82.32 | Improved |
| add SFT balanced train scheme (ablation) | Baseline face network · Long-tailed MS-Celeb-1M subset (head classes) | accuracy (%): 95.61 -> 95.77 | No clear effect |
| add SFT with accurate class center estimation (ablation) | RLL on Cars196 · Cars196 | Recall@1: 74.2 -> 80.2 | Improved |
Closest papers by meaning
- Deep Metric Learning with Spherical Embedding (NeurIPS 2020 · similarity 0.87)
- Coded Residual Transform for Generalizable Deep Metric Learning (NeurIPS 2022 · similarity 0.85)
- Curvilinear Distance Metric Learning (NeurIPS 2019 · similarity 0.85)
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.84)
- Learning Shape-Independent Transformation via Spherical Representations for Category-Level Object Pose Estimation (ICLR 2025 · similarity 0.82)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.82)
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.
