Non-isotropy Regularization for Proxy-based Deep Metric Learning
Deep Metric Learning (DML) aims to learn representation spaces on which semantic relations can simply be expressed through predefined distance metrics. Best performing approaches commonly leverage class proxies as sample stand-ins for better convergence and generalization. However, these proxy-metho...
Main Authors: | , , |
---|---|
Format: | Report |
Language: | unknown |
Published: |
arXiv
2022
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.2203.08547 https://arxiv.org/abs/2203.08547 |
id |
ftdatacite:10.48550/arxiv.2203.08547 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.2203.08547 2023-05-15T16:02:00+02:00 Non-isotropy Regularization for Proxy-based Deep Metric Learning Roth, Karsten Vinyals, Oriol Akata, Zeynep 2022 https://dx.doi.org/10.48550/arxiv.2203.08547 https://arxiv.org/abs/2203.08547 unknown arXiv Creative Commons Attribution 4.0 International https://creativecommons.org/licenses/by/4.0/legalcode cc-by-4.0 CC-BY Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences Preprint Article article CreativeWork 2022 ftdatacite https://doi.org/10.48550/arxiv.2203.08547 2022-04-01T15:53:32Z Deep Metric Learning (DML) aims to learn representation spaces on which semantic relations can simply be expressed through predefined distance metrics. Best performing approaches commonly leverage class proxies as sample stand-ins for better convergence and generalization. However, these proxy-methods solely optimize for sample-proxy distances. Given the inherent non-bijectiveness of used distance functions, this can induce locally isotropic sample distributions, leading to crucial semantic context being missed due to difficulties resolving local structures and intraclass relations between samples. To alleviate this problem, we propose non-isotropy regularization ($\mathbb{NIR}$) for proxy-based Deep Metric Learning. By leveraging Normalizing Flows, we enforce unique translatability of samples from their respective class proxies. This allows us to explicitly induce a non-isotropic distribution of samples around a proxy to optimize for. In doing so, we equip proxy-based objectives to better learn local structures. Extensive experiments highlight consistent generalization benefits of $\mathbb{NIR}$ while achieving competitive and state-of-the-art performance on the standard benchmarks CUB200-2011, Cars196 and Stanford Online Products. In addition, we find the superior convergence properties of proxy-based methods to still be retained or even improved, making $\mathbb{NIR}$ very attractive for practical usage. Code available at https://github.com/ExplainableML/NonIsotropicProxyDML. : Accepted to CVPR 2022 Report DML DataCite Metadata Store (German National Library of Science and Technology) |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
unknown |
topic |
Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences |
spellingShingle |
Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences Roth, Karsten Vinyals, Oriol Akata, Zeynep Non-isotropy Regularization for Proxy-based Deep Metric Learning |
topic_facet |
Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences |
description |
Deep Metric Learning (DML) aims to learn representation spaces on which semantic relations can simply be expressed through predefined distance metrics. Best performing approaches commonly leverage class proxies as sample stand-ins for better convergence and generalization. However, these proxy-methods solely optimize for sample-proxy distances. Given the inherent non-bijectiveness of used distance functions, this can induce locally isotropic sample distributions, leading to crucial semantic context being missed due to difficulties resolving local structures and intraclass relations between samples. To alleviate this problem, we propose non-isotropy regularization ($\mathbb{NIR}$) for proxy-based Deep Metric Learning. By leveraging Normalizing Flows, we enforce unique translatability of samples from their respective class proxies. This allows us to explicitly induce a non-isotropic distribution of samples around a proxy to optimize for. In doing so, we equip proxy-based objectives to better learn local structures. Extensive experiments highlight consistent generalization benefits of $\mathbb{NIR}$ while achieving competitive and state-of-the-art performance on the standard benchmarks CUB200-2011, Cars196 and Stanford Online Products. In addition, we find the superior convergence properties of proxy-based methods to still be retained or even improved, making $\mathbb{NIR}$ very attractive for practical usage. Code available at https://github.com/ExplainableML/NonIsotropicProxyDML. : Accepted to CVPR 2022 |
format |
Report |
author |
Roth, Karsten Vinyals, Oriol Akata, Zeynep |
author_facet |
Roth, Karsten Vinyals, Oriol Akata, Zeynep |
author_sort |
Roth, Karsten |
title |
Non-isotropy Regularization for Proxy-based Deep Metric Learning |
title_short |
Non-isotropy Regularization for Proxy-based Deep Metric Learning |
title_full |
Non-isotropy Regularization for Proxy-based Deep Metric Learning |
title_fullStr |
Non-isotropy Regularization for Proxy-based Deep Metric Learning |
title_full_unstemmed |
Non-isotropy Regularization for Proxy-based Deep Metric Learning |
title_sort |
non-isotropy regularization for proxy-based deep metric learning |
publisher |
arXiv |
publishDate |
2022 |
url |
https://dx.doi.org/10.48550/arxiv.2203.08547 https://arxiv.org/abs/2203.08547 |
genre |
DML |
genre_facet |
DML |
op_rights |
Creative Commons Attribution 4.0 International https://creativecommons.org/licenses/by/4.0/legalcode cc-by-4.0 |
op_rightsnorm |
CC-BY |
op_doi |
https://doi.org/10.48550/arxiv.2203.08547 |
_version_ |
1766397652299153408 |