Deep metric learning and image classification with nearest neighbour gaussian kernels

Benjamin J. Meyer, Ben Harwood, Tom Drummond

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

3 Citations (Scopus)

Abstract

We present a Gaussian kernel loss function and training algorithm for convolutional neural networks that can be directly applied to both distance metric learning and image classification problems. Our method treats all training features from a deep neural network as Gaussian kernel centres and computes loss by summing the influence of a feature's nearby centres in the feature embedding space. Our approach is made scalable by treating it as an approximate nearest neighbour search problem. We show how to make end-to-end learning feasible, resulting in a well formed embedding space, in which semantically related instances are likely to be located near one another, regardless of whether or not the network was trained on those classes. Our approach outperforms state-of-the-art deep metric learning approaches on embedding learning challenges, as well as conventional softmax classification on several datasets.

Original languageEnglish
Title of host publication2018 IEEE International Conference on Image Processing - Proceedings
EditorsNikolaos Boulgouris, Lisimachos P. Kondi
Place of PublicationPiscataway NJ USA
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages151-155
Number of pages5
ISBN (Electronic)9781479970612
ISBN (Print)9781479970629
DOIs
Publication statusPublished - 2018
EventIEEE International Conference on Image Processing 2018 - Athens, Greece
Duration: 7 Oct 201810 Oct 2018
Conference number: 25th
https://2018.ieeeicip.org/

Conference

ConferenceIEEE International Conference on Image Processing 2018
Abbreviated titleICIP 2018
CountryGreece
CityAthens
Period7/10/1810/10/18
Internet address

Keywords

  • Deep Learning
  • Gaussian Kernel
  • Image Classification
  • Metric Learning
  • Transfer Learning

Cite this

Meyer, B. J., Harwood, B., & Drummond, T. (2018). Deep metric learning and image classification with nearest neighbour gaussian kernels. In N. Boulgouris, & L. P. Kondi (Eds.), 2018 IEEE International Conference on Image Processing - Proceedings (pp. 151-155). Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICIP.2018.8451297
Meyer, Benjamin J. ; Harwood, Ben ; Drummond, Tom. / Deep metric learning and image classification with nearest neighbour gaussian kernels. 2018 IEEE International Conference on Image Processing - Proceedings. editor / Nikolaos Boulgouris ; Lisimachos P. Kondi. Piscataway NJ USA : IEEE, Institute of Electrical and Electronics Engineers, 2018. pp. 151-155
@inproceedings{d6f9488566ab44db86b0992880822387,
title = "Deep metric learning and image classification with nearest neighbour gaussian kernels",
abstract = "We present a Gaussian kernel loss function and training algorithm for convolutional neural networks that can be directly applied to both distance metric learning and image classification problems. Our method treats all training features from a deep neural network as Gaussian kernel centres and computes loss by summing the influence of a feature's nearby centres in the feature embedding space. Our approach is made scalable by treating it as an approximate nearest neighbour search problem. We show how to make end-to-end learning feasible, resulting in a well formed embedding space, in which semantically related instances are likely to be located near one another, regardless of whether or not the network was trained on those classes. Our approach outperforms state-of-the-art deep metric learning approaches on embedding learning challenges, as well as conventional softmax classification on several datasets.",
keywords = "Deep Learning, Gaussian Kernel, Image Classification, Metric Learning, Transfer Learning",
author = "Meyer, {Benjamin J.} and Ben Harwood and Tom Drummond",
year = "2018",
doi = "10.1109/ICIP.2018.8451297",
language = "English",
isbn = "9781479970629",
pages = "151--155",
editor = "Nikolaos Boulgouris and Kondi, {Lisimachos P.}",
booktitle = "2018 IEEE International Conference on Image Processing - Proceedings",
publisher = "IEEE, Institute of Electrical and Electronics Engineers",
address = "United States of America",

}

Meyer, BJ, Harwood, B & Drummond, T 2018, Deep metric learning and image classification with nearest neighbour gaussian kernels. in N Boulgouris & LP Kondi (eds), 2018 IEEE International Conference on Image Processing - Proceedings. IEEE, Institute of Electrical and Electronics Engineers, Piscataway NJ USA, pp. 151-155, IEEE International Conference on Image Processing 2018, Athens, Greece, 7/10/18. https://doi.org/10.1109/ICIP.2018.8451297

Deep metric learning and image classification with nearest neighbour gaussian kernels. / Meyer, Benjamin J.; Harwood, Ben; Drummond, Tom.

2018 IEEE International Conference on Image Processing - Proceedings. ed. / Nikolaos Boulgouris; Lisimachos P. Kondi. Piscataway NJ USA : IEEE, Institute of Electrical and Electronics Engineers, 2018. p. 151-155.

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

TY - GEN

T1 - Deep metric learning and image classification with nearest neighbour gaussian kernels

AU - Meyer, Benjamin J.

AU - Harwood, Ben

AU - Drummond, Tom

PY - 2018

Y1 - 2018

N2 - We present a Gaussian kernel loss function and training algorithm for convolutional neural networks that can be directly applied to both distance metric learning and image classification problems. Our method treats all training features from a deep neural network as Gaussian kernel centres and computes loss by summing the influence of a feature's nearby centres in the feature embedding space. Our approach is made scalable by treating it as an approximate nearest neighbour search problem. We show how to make end-to-end learning feasible, resulting in a well formed embedding space, in which semantically related instances are likely to be located near one another, regardless of whether or not the network was trained on those classes. Our approach outperforms state-of-the-art deep metric learning approaches on embedding learning challenges, as well as conventional softmax classification on several datasets.

AB - We present a Gaussian kernel loss function and training algorithm for convolutional neural networks that can be directly applied to both distance metric learning and image classification problems. Our method treats all training features from a deep neural network as Gaussian kernel centres and computes loss by summing the influence of a feature's nearby centres in the feature embedding space. Our approach is made scalable by treating it as an approximate nearest neighbour search problem. We show how to make end-to-end learning feasible, resulting in a well formed embedding space, in which semantically related instances are likely to be located near one another, regardless of whether or not the network was trained on those classes. Our approach outperforms state-of-the-art deep metric learning approaches on embedding learning challenges, as well as conventional softmax classification on several datasets.

KW - Deep Learning

KW - Gaussian Kernel

KW - Image Classification

KW - Metric Learning

KW - Transfer Learning

UR - http://www.scopus.com/inward/record.url?scp=85062900921&partnerID=8YFLogxK

U2 - 10.1109/ICIP.2018.8451297

DO - 10.1109/ICIP.2018.8451297

M3 - Conference Paper

AN - SCOPUS:85062900921

SN - 9781479970629

SP - 151

EP - 155

BT - 2018 IEEE International Conference on Image Processing - Proceedings

A2 - Boulgouris, Nikolaos

A2 - Kondi, Lisimachos P.

PB - IEEE, Institute of Electrical and Electronics Engineers

CY - Piscataway NJ USA

ER -

Meyer BJ, Harwood B, Drummond T. Deep metric learning and image classification with nearest neighbour gaussian kernels. In Boulgouris N, Kondi LP, editors, 2018 IEEE International Conference on Image Processing - Proceedings. Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers. 2018. p. 151-155 https://doi.org/10.1109/ICIP.2018.8451297