Modelling local deep convolutional neural network features to improve fine-grained image classification

Zongyuan Ge, Chris McCool, Conrad Sanderson, Peter Corke

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

Abstract

We propose a local modelling approach using deep convolutional neural networks (CNNs) for fine-grained image classification. Recently, deep CNNs trained from large datasets have considerably improved the performance of object recognition. However, to date there has been limited work using these deep CNNs as local feature extractors. This partly stems from CNNs having internal representations which are high dimensional, thereby making such representations difficult to model using stochastic models. To overcome this issue, we propose to reduce the dimensionality of one of the internal fully connected layers, in conjunction with layer-restricted retraining to avoid retraining the entire network. The distribution of low-dimensional features obtained from the modified layer is then modelled using a Gaussian mixture model. Comparative experiments show that considerable performance improvements can be achieved on the challenging Fish and UEC FOOD-100 datasets.

Original languageEnglish
Title of host publication2015 IEEE International Conference on Image Processing (ICIP 2015)
EditorsFabrice Labeau, Jean-Philippe Thiran
Place of PublicationPiscataway NJ USA
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages4112-4116
Number of pages5
ISBN (Electronic)9781479983391
ISBN (Print)9781479983407
DOIs
Publication statusPublished - 9 Dec 2015
Externally publishedYes
EventIEEE International Conference on Image Processing 2015 - Quebec City, Canada
Duration: 27 Sep 201530 Sep 2015
Conference number: 22nd
http://icip2015.org/index.html

Conference

ConferenceIEEE International Conference on Image Processing 2015
Abbreviated titleICIP 2015
CountryCanada
CityQuebec City
Period27/09/1530/09/15
Internet address

Keywords

  • deep convolutional neural networks
  • fine-grained classification
  • Gaussian mixture models
  • session variation modelling

Cite this

Ge, Z., McCool, C., Sanderson, C., & Corke, P. (2015). Modelling local deep convolutional neural network features to improve fine-grained image classification. In F. Labeau, & J-P. Thiran (Eds.), 2015 IEEE International Conference on Image Processing (ICIP 2015) (pp. 4112-4116). [7351579] Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICIP.2015.7351579
Ge, Zongyuan ; McCool, Chris ; Sanderson, Conrad ; Corke, Peter. / Modelling local deep convolutional neural network features to improve fine-grained image classification. 2015 IEEE International Conference on Image Processing (ICIP 2015). editor / Fabrice Labeau ; Jean-Philippe Thiran. Piscataway NJ USA : IEEE, Institute of Electrical and Electronics Engineers, 2015. pp. 4112-4116
@inproceedings{d5da31f5b8a84c20a0db3f42a2486130,
title = "Modelling local deep convolutional neural network features to improve fine-grained image classification",
abstract = "We propose a local modelling approach using deep convolutional neural networks (CNNs) for fine-grained image classification. Recently, deep CNNs trained from large datasets have considerably improved the performance of object recognition. However, to date there has been limited work using these deep CNNs as local feature extractors. This partly stems from CNNs having internal representations which are high dimensional, thereby making such representations difficult to model using stochastic models. To overcome this issue, we propose to reduce the dimensionality of one of the internal fully connected layers, in conjunction with layer-restricted retraining to avoid retraining the entire network. The distribution of low-dimensional features obtained from the modified layer is then modelled using a Gaussian mixture model. Comparative experiments show that considerable performance improvements can be achieved on the challenging Fish and UEC FOOD-100 datasets.",
keywords = "deep convolutional neural networks, fine-grained classification, Gaussian mixture models, session variation modelling",
author = "Zongyuan Ge and Chris McCool and Conrad Sanderson and Peter Corke",
year = "2015",
month = "12",
day = "9",
doi = "10.1109/ICIP.2015.7351579",
language = "English",
isbn = "9781479983407",
pages = "4112--4116",
editor = "Fabrice Labeau and Jean-Philippe Thiran",
booktitle = "2015 IEEE International Conference on Image Processing (ICIP 2015)",
publisher = "IEEE, Institute of Electrical and Electronics Engineers",
address = "United States of America",

}

Ge, Z, McCool, C, Sanderson, C & Corke, P 2015, Modelling local deep convolutional neural network features to improve fine-grained image classification. in F Labeau & J-P Thiran (eds), 2015 IEEE International Conference on Image Processing (ICIP 2015)., 7351579, IEEE, Institute of Electrical and Electronics Engineers, Piscataway NJ USA, pp. 4112-4116, IEEE International Conference on Image Processing 2015, Quebec City, Canada, 27/09/15. https://doi.org/10.1109/ICIP.2015.7351579

Modelling local deep convolutional neural network features to improve fine-grained image classification. / Ge, Zongyuan; McCool, Chris; Sanderson, Conrad; Corke, Peter.

2015 IEEE International Conference on Image Processing (ICIP 2015). ed. / Fabrice Labeau; Jean-Philippe Thiran. Piscataway NJ USA : IEEE, Institute of Electrical and Electronics Engineers, 2015. p. 4112-4116 7351579.

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

TY - GEN

T1 - Modelling local deep convolutional neural network features to improve fine-grained image classification

AU - Ge, Zongyuan

AU - McCool, Chris

AU - Sanderson, Conrad

AU - Corke, Peter

PY - 2015/12/9

Y1 - 2015/12/9

N2 - We propose a local modelling approach using deep convolutional neural networks (CNNs) for fine-grained image classification. Recently, deep CNNs trained from large datasets have considerably improved the performance of object recognition. However, to date there has been limited work using these deep CNNs as local feature extractors. This partly stems from CNNs having internal representations which are high dimensional, thereby making such representations difficult to model using stochastic models. To overcome this issue, we propose to reduce the dimensionality of one of the internal fully connected layers, in conjunction with layer-restricted retraining to avoid retraining the entire network. The distribution of low-dimensional features obtained from the modified layer is then modelled using a Gaussian mixture model. Comparative experiments show that considerable performance improvements can be achieved on the challenging Fish and UEC FOOD-100 datasets.

AB - We propose a local modelling approach using deep convolutional neural networks (CNNs) for fine-grained image classification. Recently, deep CNNs trained from large datasets have considerably improved the performance of object recognition. However, to date there has been limited work using these deep CNNs as local feature extractors. This partly stems from CNNs having internal representations which are high dimensional, thereby making such representations difficult to model using stochastic models. To overcome this issue, we propose to reduce the dimensionality of one of the internal fully connected layers, in conjunction with layer-restricted retraining to avoid retraining the entire network. The distribution of low-dimensional features obtained from the modified layer is then modelled using a Gaussian mixture model. Comparative experiments show that considerable performance improvements can be achieved on the challenging Fish and UEC FOOD-100 datasets.

KW - deep convolutional neural networks

KW - fine-grained classification

KW - Gaussian mixture models

KW - session variation modelling

UR - http://www.scopus.com/inward/record.url?scp=84956616783&partnerID=8YFLogxK

U2 - 10.1109/ICIP.2015.7351579

DO - 10.1109/ICIP.2015.7351579

M3 - Conference Paper

SN - 9781479983407

SP - 4112

EP - 4116

BT - 2015 IEEE International Conference on Image Processing (ICIP 2015)

A2 - Labeau, Fabrice

A2 - Thiran, Jean-Philippe

PB - IEEE, Institute of Electrical and Electronics Engineers

CY - Piscataway NJ USA

ER -

Ge Z, McCool C, Sanderson C, Corke P. Modelling local deep convolutional neural network features to improve fine-grained image classification. In Labeau F, Thiran J-P, editors, 2015 IEEE International Conference on Image Processing (ICIP 2015). Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers. 2015. p. 4112-4116. 7351579 https://doi.org/10.1109/ICIP.2015.7351579