Research map / Label-efficient and robust learning / Recognition and long-tail
A Non-Isotropic Probabilistic Take On Proxy-Based Deep Metric Learning
Michael Kirchhof, Karsten Roth, Zeynep Akata, Enkelejda Kasneci
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 |
|---|---|---|---|
| make ProxyNCA probabilistic (EL-nivMF) | ProxyNCA · CUB200-2011 | R@1: 63.2 -> 64.8 | Improved |
| make ProxyNCA probabilistic (EL-nivMF) | ProxyNCA · CARS196 | R@1: 78.8 -> 82.1 | Improved |
| make ProxyNCA probabilistic (EL-nivMF) | ProxyNCA · SOP | R@1: 76.2 -> 76.6 | Improved |
| add EL-nivMF as regularizer on ProxyAnchor | ProxyAnchor · CUB200-2011 | R@1: 64.4 -> 66.5 | Improved |
| add EL-nivMF as regularizer on ProxyAnchor | ProxyAnchor · CARS196 | R@1: 82.4 -> 83.6 | Improved |
| add EL-nivMF as regularizer on ProxyAnchor | ProxyAnchor · SOP | R@1: 78.0 -> 78.2 | Improved |
| add non-isotropy to cosine distance (d_nivMF) (ablation) | proxy DML with d_Cos · CUB200-2011 | R@1: +2.1pp | Improved |
| add non-isotropy to cosine distance (d_nivMF) (ablation) | proxy DML with d_Cos · CARS196 | R@1: +1.7pp | Improved |
| additionally consider image norm (d_EL-nivMF over d_nivMF) (ablation) | non-isotropic proxy DML · CUB200-2011 | R@1: +0.6pp | Improved |
| additionally consider image norm (d_EL-nivMF over d_nivMF) (ablation) | non-isotropic proxy DML · CARS196 | R@1: +0.3pp | Improved |
| use L2 distance instead of cosine distance as retrieval metric (ablation) | models trained with various distances · CUB200-2011 and CARS196 | R@1: -5.34pp | Hurt |
| drop the normalizer D(K) in the nivMF density (ablation) | nivMF · empirical tests | performance | Hurt |
Closest papers by meaning
- Non-Isotropy Regularization for Proxy-Based Deep Metric Learning (CVPR 2022 · similarity 0.92)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.91)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.88)
- Potential Field Based Deep Metric Learning (CVPR 2025 · similarity 0.88)
- Learning Semantic Proxies from Visual Prompts for Parameter-Efficient Fine-Tuning in Deep Metric Learning (ICLR 2024 · similarity 0.87)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · 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.
