ProxyDR: Deep Hyperspherical Metric Learning with Distance Ratio-Based Formulation
Abstract
In this paper, we propose ProxyDR, a novel metric learning method for hyperspherical embeddings. Through the adoption of a distance ratio-based formulation, ProxyDR resolves the fundamental shortcomings of the conventional squared distance softmax formulation. Notably, our proposed method addresses the near-uniform positioning of class representatives that obstructs effective learning of semantic relationships among classes—a phenomenon demonstrated by our theoretical and experimental analyses. Moreover, by employing proxies as class representatives, our method can be effortlessly incorporated into established classification frameworks. We rigorously evaluate ProxyDR against conventional methods using diverse datasets, including CIFAR100 and NABirds, demonstrating superiority in capturing hierarchical structures while maintaining conventional classification accuracy.
Cite
Text
Kim et al. "ProxyDR: Deep Hyperspherical Metric Learning with Distance Ratio-Based Formulation." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91585-7_26Markdown
[Kim et al. "ProxyDR: Deep Hyperspherical Metric Learning with Distance Ratio-Based Formulation." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/kim2024eccvw-proxydr/) doi:10.1007/978-3-031-91585-7_26BibTeX
@inproceedings{kim2024eccvw-proxydr,
title = {{ProxyDR: Deep Hyperspherical Metric Learning with Distance Ratio-Based Formulation}},
author = {Kim, Hyeongji and Choi, Changkyu and Kampffmeyer, Michael and Berge, Terje and Parviainen, Pekka and Malde, Ketil},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {434-450},
doi = {10.1007/978-3-031-91585-7_26},
url = {https://mlanthology.org/eccvw/2024/kim2024eccvw-proxydr/}
}