Adaptive Metric Learning for People Re-Identification

Guanwen ZHANG; Jien KATO; Yu WANG; Kenji MASE

doi:10.1587/transinf.2013EDP7451

Adaptive Metric Learning for People Re-Identification

Guanwen ZHANG, Jien KATO, Yu WANG, Kenji MASE

Full Text Views

0

Cite this

Summary :

There exist two intrinsic issues in multiple-shot person re-identification: (1) large differences in camera view, illumination, and non-rigid deformation of posture that make the intra-class variance even larger than the inter-class variance; (2) only a few training data that are available for learning tasks in a realistic re-identification scenario. In our previous work, we proposed a local distance comparison framework to deal with the first issue. In this paper, to deal with the second issue (i.e., to derive a reliable distance metric from limited training data), we propose an adaptive learning method to learn an adaptive distance metric, which integrates prior knowledge learned from a large existing auxiliary dataset and task-specific information extracted from a much smaller training dataset. Experimental results on several public benchmark datasets show that combined with the local distance comparison framework, our adaptive learning method is superior to conventional approaches.

Publication: IEICE TRANSACTIONS on Information Vol.E97-D No.11 pp.2888-2902

Publication Date: 2014/11/01

Publicized

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2013EDP7451

Type of Manuscript: PAPER

Category: Image Processing and Video Processing

Authors

Guanwen ZHANG
  Nagoya University
Jien KATO
  Nagoya University
Yu WANG
  Nagoya University
Kenji MASE
  Nagoya University

Keyword

multiple-shot people re-identification, adaptive metric learning, local distance comparison

Cite this

Copy

Guanwen ZHANG, Jien KATO, Yu WANG, Kenji MASE, "Adaptive Metric Learning for People Re-Identification" in IEICE TRANSACTIONS on Information, vol. E97-D, no. 11, pp. 2888-2902, November 2014, doi: 10.1587/transinf.2013EDP7451.
Abstract: There exist two intrinsic issues in multiple-shot person re-identification: (1) large differences in camera view, illumination, and non-rigid deformation of posture that make the intra-class variance even larger than the inter-class variance; (2) only a few training data that are available for learning tasks in a realistic re-identification scenario. In our previous work, we proposed a local distance comparison framework to deal with the first issue. In this paper, to deal with the second issue (i.e., to derive a reliable distance metric from limited training data), we propose an adaptive learning method to learn an adaptive distance metric, which integrates prior knowledge learned from a large existing auxiliary dataset and task-specific information extracted from a much smaller training dataset. Experimental results on several public benchmark datasets show that combined with the local distance comparison framework, our adaptive learning method is superior to conventional approaches.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2013EDP7451/_p

Copy

@ARTICLE{e97-d_11_2888,
author={Guanwen ZHANG, Jien KATO, Yu WANG, Kenji MASE, },
journal={IEICE TRANSACTIONS on Information},
title={Adaptive Metric Learning for People Re-Identification},
year={2014},
volume={E97-D},
number={11},
pages={2888-2902},
abstract={There exist two intrinsic issues in multiple-shot person re-identification: (1) large differences in camera view, illumination, and non-rigid deformation of posture that make the intra-class variance even larger than the inter-class variance; (2) only a few training data that are available for learning tasks in a realistic re-identification scenario. In our previous work, we proposed a local distance comparison framework to deal with the first issue. In this paper, to deal with the second issue (i.e., to derive a reliable distance metric from limited training data), we propose an adaptive learning method to learn an adaptive distance metric, which integrates prior knowledge learned from a large existing auxiliary dataset and task-specific information extracted from a much smaller training dataset. Experimental results on several public benchmark datasets show that combined with the local distance comparison framework, our adaptive learning method is superior to conventional approaches.},
keywords={},
doi={10.1587/transinf.2013EDP7451},
ISSN={1745-1361},
month={November},}

Copy

TY - JOUR
TI - Adaptive Metric Learning for People Re-Identification
T2 - IEICE TRANSACTIONS on Information
SP - 2888
EP - 2902
AU - Guanwen ZHANG
AU - Jien KATO
AU - Yu WANG
AU - Kenji MASE
PY - 2014
DO - 10.1587/transinf.2013EDP7451
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E97-D
IS - 11
JA - IEICE TRANSACTIONS on Information
Y1 - November 2014
AB - There exist two intrinsic issues in multiple-shot person re-identification: (1) large differences in camera view, illumination, and non-rigid deformation of posture that make the intra-class variance even larger than the inter-class variance; (2) only a few training data that are available for learning tasks in a realistic re-identification scenario. In our previous work, we proposed a local distance comparison framework to deal with the first issue. In this paper, to deal with the second issue (i.e., to derive a reliable distance metric from limited training data), we propose an adaptive learning method to learn an adaptive distance metric, which integrates prior knowledge learned from a large existing auxiliary dataset and task-specific information extracted from a much smaller training dataset. Experimental results on several public benchmark datasets show that combined with the local distance comparison framework, our adaptive learning method is superior to conventional approaches.
ER -