Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature

Junbiao Pang; Huihuang Lin; Li Su; Chunjie Zhang; Weigang Zhang; Lijuan Duan; Qingming Huang; Baocai Yin

doi:10.1109/ICIP.2016.7532515

Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature

Junbiao Pang
, Huihuang Lin
, Li Su
, Chunjie Zhang
, Weigang Zhang
, Lijuan Duan
, Qingming Huang
, Baocai Yin

Beijing Key Laboratory of Intelligent Telecommunication Software and Multimedia
Beijing University of Technology
University of Chinese Academy of Sciences
Chinese Academy of Sciences
School of Computer Science and Technology, Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

3 Scopus citations

Abstract

Convolutional Neural Networks (CNNs) have delivered impressive state-of-the-art performances for many vision tasks, while the computation costs of these networks during test-time are notorious. Empirical results have discovered that CNNs have learned the redundant representations both within and across different layers. When CNNs are applied for binary classification, we investigate a method to exploit this redundancy across layers, and construct a cascade of classifiers which explicitly balances classification accuracy and hierarchical feature extraction costs. Our method cost-sensitively selects feature points across several layers from trained networks and embeds non-expensive yet discriminative features into a cascade. Experiments on binary classification demonstrate that our framework leads to drastic test-time improvements, e.g., possible 47.2x speedup for TRECVID upper body detection, 2.82x speedup for Pascal VOC2007 People detection, 3.72x for INRIA Person detection with less than 0.5% drop in accuracies of the original networks.

Original language	English
Title of host publication	2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings
Publisher	IEEE Computer Society
Pages	1037-1041
Number of pages	5
ISBN (Electronic)	9781467399616
DOIs	https://doi.org/10.1109/ICIP.2016.7532515
State	Published - 3 Aug 2016
Externally published	Yes
Event	23rd IEEE International Conference on Image Processing, ICIP 2016 - Phoenix, United States Duration: 25 Sep 2016 → 28 Sep 2016

Publication series

Name	Proceedings - International Conference on Image Processing, ICIP
Volume	2016-August
ISSN (Print)	1522-4880

Conference

Conference	23rd IEEE International Conference on Image Processing, ICIP 2016
Country/Territory	United States
City	Phoenix
Period	25/09/16 → 28/09/16

Keywords

Accelerate
Binary classification
Cascade
Convolutional Neural Networks
Cost-sensitive
Feature selection

Access to Document

10.1109/ICIP.2016.7532515

Cite this

Pang, J., Lin, H., Su, L., Zhang, C., Zhang, W., Duan, L., Huang, Q., & Yin, B. (2016). Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature. In 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings (pp. 1037-1041). Article 7532515 (Proceedings - International Conference on Image Processing, ICIP; Vol. 2016-August). IEEE Computer Society. https://doi.org/10.1109/ICIP.2016.7532515

@inproceedings{553d6b6e7d974ff597b6a77639bbc51e,

title = "Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature",

abstract = "Convolutional Neural Networks (CNNs) have delivered impressive state-of-the-art performances for many vision tasks, while the computation costs of these networks during test-time are notorious. Empirical results have discovered that CNNs have learned the redundant representations both within and across different layers. When CNNs are applied for binary classification, we investigate a method to exploit this redundancy across layers, and construct a cascade of classifiers which explicitly balances classification accuracy and hierarchical feature extraction costs. Our method cost-sensitively selects feature points across several layers from trained networks and embeds non-expensive yet discriminative features into a cascade. Experiments on binary classification demonstrate that our framework leads to drastic test-time improvements, e.g., possible 47.2x speedup for TRECVID upper body detection, 2.82x speedup for Pascal VOC2007 People detection, 3.72x for INRIA Person detection with less than 0.5\% drop in accuracies of the original networks.",

keywords = "Accelerate, Binary classification, Cascade, Convolutional Neural Networks, Cost-sensitive, Feature selection",

author = "Junbiao Pang and Huihuang Lin and Li Su and Chunjie Zhang and Weigang Zhang and Lijuan Duan and Qingming Huang and Baocai Yin",

note = "Publisher Copyright: {\textcopyright} 2016 IEEE.; 23rd IEEE International Conference on Image Processing, ICIP 2016 ; Conference date: 25-09-2016 Through 28-09-2016",

year = "2016",

month = aug,

day = "3",

doi = "10.1109/ICIP.2016.7532515",

language = "英语",

series = "Proceedings - International Conference on Image Processing, ICIP",

publisher = "IEEE Computer Society",

pages = "1037--1041",

booktitle = "2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings",

address = "美国",

}

Pang, J, Lin, H, Su, L, Zhang, C, Zhang, W, Duan, L, Huang, Q & Yin, B 2016, Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature. in 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings., 7532515, Proceedings - International Conference on Image Processing, ICIP, vol. 2016-August, IEEE Computer Society, pp. 1037-1041, 23rd IEEE International Conference on Image Processing, ICIP 2016, Phoenix, United States, 25/09/16. https://doi.org/10.1109/ICIP.2016.7532515

Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature. / Pang, Junbiao; Lin, Huihuang; Su, Li et al.
2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings. IEEE Computer Society, 2016. p. 1037-1041 7532515 (Proceedings - International Conference on Image Processing, ICIP; Vol. 2016-August).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature

AU - Pang, Junbiao

AU - Lin, Huihuang

AU - Su, Li

AU - Zhang, Chunjie

AU - Zhang, Weigang

AU - Duan, Lijuan

AU - Huang, Qingming

AU - Yin, Baocai

PY - 2016/8/3

Y1 - 2016/8/3

N2 - Convolutional Neural Networks (CNNs) have delivered impressive state-of-the-art performances for many vision tasks, while the computation costs of these networks during test-time are notorious. Empirical results have discovered that CNNs have learned the redundant representations both within and across different layers. When CNNs are applied for binary classification, we investigate a method to exploit this redundancy across layers, and construct a cascade of classifiers which explicitly balances classification accuracy and hierarchical feature extraction costs. Our method cost-sensitively selects feature points across several layers from trained networks and embeds non-expensive yet discriminative features into a cascade. Experiments on binary classification demonstrate that our framework leads to drastic test-time improvements, e.g., possible 47.2x speedup for TRECVID upper body detection, 2.82x speedup for Pascal VOC2007 People detection, 3.72x for INRIA Person detection with less than 0.5% drop in accuracies of the original networks.

AB - Convolutional Neural Networks (CNNs) have delivered impressive state-of-the-art performances for many vision tasks, while the computation costs of these networks during test-time are notorious. Empirical results have discovered that CNNs have learned the redundant representations both within and across different layers. When CNNs are applied for binary classification, we investigate a method to exploit this redundancy across layers, and construct a cascade of classifiers which explicitly balances classification accuracy and hierarchical feature extraction costs. Our method cost-sensitively selects feature points across several layers from trained networks and embeds non-expensive yet discriminative features into a cascade. Experiments on binary classification demonstrate that our framework leads to drastic test-time improvements, e.g., possible 47.2x speedup for TRECVID upper body detection, 2.82x speedup for Pascal VOC2007 People detection, 3.72x for INRIA Person detection with less than 0.5% drop in accuracies of the original networks.

KW - Accelerate

KW - Binary classification

KW - Cascade

KW - Convolutional Neural Networks

KW - Cost-sensitive

KW - Feature selection

UR - https://www.scopus.com/pages/publications/85006810292

U2 - 10.1109/ICIP.2016.7532515

DO - 10.1109/ICIP.2016.7532515

M3 - 会议稿件

AN - SCOPUS:85006810292

T3 - Proceedings - International Conference on Image Processing, ICIP

SP - 1037

EP - 1041

BT - 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings

PB - IEEE Computer Society

T2 - 23rd IEEE International Conference on Image Processing, ICIP 2016

Y2 - 25 September 2016 through 28 September 2016

ER -

Pang J, Lin H, Su L, Zhang C, Zhang W, Duan L et al. Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature. In 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings. IEEE Computer Society. 2016. p. 1037-1041. 7532515. (Proceedings - International Conference on Image Processing, ICIP). doi: 10.1109/ICIP.2016.7532515

Accelerate convolutional neural networks for binary classification via cascading cost-sensitive feature

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this