Cross-Batch Memory for Embedding Learning
Mining informative negative instances are of central importance to deep metric learning (DML), however this task is intrinsically limited by mini-batch training, where only a mini-batch of instances is accessible at each iteration. In this paper, we identify a "slow drift" phenomena by obs...
Main Authors: | , , , |
---|---|
Format: | Article in Journal/Newspaper |
Language: | unknown |
Published: |
arXiv
2019
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.1912.06798 https://arxiv.org/abs/1912.06798 |
id |
ftdatacite:10.48550/arxiv.1912.06798 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.1912.06798 2023-05-15T16:01:31+02:00 Cross-Batch Memory for Embedding Learning Wang, Xun Zhang, Haozhi Huang, Weilin Scott, Matthew R. 2019 https://dx.doi.org/10.48550/arxiv.1912.06798 https://arxiv.org/abs/1912.06798 unknown arXiv arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ Machine Learning cs.LG Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences Article CreativeWork article Preprint 2019 ftdatacite https://doi.org/10.48550/arxiv.1912.06798 2022-03-10T16:32:42Z Mining informative negative instances are of central importance to deep metric learning (DML), however this task is intrinsically limited by mini-batch training, where only a mini-batch of instances is accessible at each iteration. In this paper, we identify a "slow drift" phenomena by observing that the embedding features drift exceptionally slow even as the model parameters are updating throughout the training process. This suggests that the features of instances computed at preceding iterations can be used to considerably approximate their features extracted by the current model. We propose a cross-batch memory (XBM) mechanism that memorizes the embeddings of past iterations, allowing the model to collect sufficient hard negative pairs across multiple mini-batches - even over the whole dataset. Our XBM can be directly integrated into a general pair-based DML framework, where the XBM augmented DML can boost performance considerably. In particular, without bells and whistles, a simple contrastive loss with our XBM can have large R@1 improvements of 12%-22.5% on three large-scale image retrieval datasets, surpassing the most sophisticated state-of-the-art methods, by a large margin. Our XBM is conceptually simple, easy to implement - using several lines of codes, and is memory efficient - with a negligible 0.2 GB extra GPU memory. Code is available at: https://github.com/MalongTech/research-xbm. : CVPR 2020 Oral Article in Journal/Newspaper 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 |
Machine Learning cs.LG Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences |
spellingShingle |
Machine Learning cs.LG Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences Wang, Xun Zhang, Haozhi Huang, Weilin Scott, Matthew R. Cross-Batch Memory for Embedding Learning |
topic_facet |
Machine Learning cs.LG Computer Vision and Pattern Recognition cs.CV FOS Computer and information sciences |
description |
Mining informative negative instances are of central importance to deep metric learning (DML), however this task is intrinsically limited by mini-batch training, where only a mini-batch of instances is accessible at each iteration. In this paper, we identify a "slow drift" phenomena by observing that the embedding features drift exceptionally slow even as the model parameters are updating throughout the training process. This suggests that the features of instances computed at preceding iterations can be used to considerably approximate their features extracted by the current model. We propose a cross-batch memory (XBM) mechanism that memorizes the embeddings of past iterations, allowing the model to collect sufficient hard negative pairs across multiple mini-batches - even over the whole dataset. Our XBM can be directly integrated into a general pair-based DML framework, where the XBM augmented DML can boost performance considerably. In particular, without bells and whistles, a simple contrastive loss with our XBM can have large R@1 improvements of 12%-22.5% on three large-scale image retrieval datasets, surpassing the most sophisticated state-of-the-art methods, by a large margin. Our XBM is conceptually simple, easy to implement - using several lines of codes, and is memory efficient - with a negligible 0.2 GB extra GPU memory. Code is available at: https://github.com/MalongTech/research-xbm. : CVPR 2020 Oral |
format |
Article in Journal/Newspaper |
author |
Wang, Xun Zhang, Haozhi Huang, Weilin Scott, Matthew R. |
author_facet |
Wang, Xun Zhang, Haozhi Huang, Weilin Scott, Matthew R. |
author_sort |
Wang, Xun |
title |
Cross-Batch Memory for Embedding Learning |
title_short |
Cross-Batch Memory for Embedding Learning |
title_full |
Cross-Batch Memory for Embedding Learning |
title_fullStr |
Cross-Batch Memory for Embedding Learning |
title_full_unstemmed |
Cross-Batch Memory for Embedding Learning |
title_sort |
cross-batch memory for embedding learning |
publisher |
arXiv |
publishDate |
2019 |
url |
https://dx.doi.org/10.48550/arxiv.1912.06798 https://arxiv.org/abs/1912.06798 |
genre |
DML |
genre_facet |
DML |
op_rights |
arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
op_doi |
https://doi.org/10.48550/arxiv.1912.06798 |
_version_ |
1766397343039488000 |