Research map / Label-efficient and robust learning / Recognition and long-tail
No Fuss Distance Metric Learning Using Proxies
Yair Movshovitz-Attias, Alexander Toshev, Thomas K. Leung, Sergey Ioffe, Saurabh Singh
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 |
|---|---|---|---|
| replace Struct Clust with Proxy-NCA loss | Proxy-NCA · Cars196 | Recall@1: 58.11 -> 73.22 | Improved |
| replace Struct Clust with Proxy-NCA loss | Proxy-NCA · Cars196 | NMI: 59.04 -> 64.90 | Improved |
| replace Struct Clust with Proxy-NCA loss | Proxy-NCA · CUB200-2011 | Recall@1: 48.18 -> 49.21 | Improved |
| replace Struct Clust with Proxy-NCA loss | Proxy-NCA · CUB200-2011 | Recall@8: 81.92 -> 72.40 | Hurt |
| replace Struct Clust with Proxy-NCA loss | Proxy-NCA · Stanford Online Products | Recall@1: over 6% gap | Improved |
| replace top method with Proxy-NCA loss | Proxy-NCA · Stanford Online Products | NMI: 89.48 -> 90.6 | Improved |
| use Proxy-NCA loss instead of Proxy-Triplet (ablation) | Proxy-Triplet · Cars196 | Recall@1: 55.90 -> 73.22 | Improved |
| add proxies to triplet loss (Triplet Semihard to Proxy-Triplet) (ablation) | Triplet Semihard · Cars196 | Recall@1: 51.54 -> 55.90 | Improved |
| use proxies in training (Proxy-NCA vs baselines) for speed | Proxy-NCA · Cars196 | convergence speed: about three times as fast | Improved |
| reduce proxies below one per class (fractional assignment) (ablation) | Proxy-NCA static · Cars196 | Recall@1: proxy-per-class >= 0.5 beats previous methods | Mixed |
| use dynamic proxy assignment with 0.5 proxies per concept (ablation) | Proxy-NCA dynamic · Cars196 | Recall@1: 0.5 proxies per concept enough to beat state of the art | Improved |
| add more proxies than semantic concepts in dynamic assignment (ablation) | Proxy-NCA dynamic · Cars196 | Recall@1 | Mixed |
Closest papers by meaning
- Hard negative examples are hard, but useful (ECCV 2020 · similarity 0.88)
- Proxy Anchor Loss for Deep Metric Learning (CVPR 2020 · similarity 0.88)
- Neural Bregman Divergences for Distance Learning (ICLR 2023 · similarity 0.88)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.87)
- HIER: Metric Learning Beyond Class Labels via Hierarchical Regularization (CVPR 2023 · similarity 0.87)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.85)
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.
