Improved object categorization and detection using comparative object similarity

Research output: Contribution to journalArticle

Abstract

Due to the intrinsic long-tailed distribution of objects in the real world, we are unlikely to be able to train an object recognizer/detector with many visual examples for each category. We have to share visual knowledge between object categories to enable learning with few or no training examples. In this paper, we show that local object similarity information - statements that pairs of categories are similar or dissimilar - is a very useful cue to tie different categories to each other for effective knowledge transfer. The key insight: Given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. To exploit this category-dependent similarity regularization, we develop a regularized kernel machine algorithm to train kernel classifiers for categories with few or no training examples. We also adapt the state-of-the-art object detector to encode object similarity constraints. Our experiments on hundreds of categories from the Labelme dataset show that our regularized kernel classifiers can make significant improvement on object categorization. We also evaluate the improved object detector on the PASCAL VOC 2007 benchmark dataset.

Original languageEnglish (US)
Article number6482138
Pages (from-to)2442-2453
Number of pages12
JournalIEEE transactions on pattern analysis and machine intelligence
Volume35
Issue number10
DOIs
StatePublished - Sep 3 2013

Fingerprint

Categorization
Detectors
Classifiers
Volatile organic compounds
Detector
Similarity
Object
Classifier
Kernel Machines
Experiments
kernel
Knowledge Transfer
Object Model
Tie
Regularization
Benchmark
Dependent
Evaluate

Keywords

  • Comparative object similarity
  • PASCAL VOC
  • SVM
  • deformable part model
  • kernel machines
  • object categorization
  • object detection
  • sharing

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition
  • Computational Theory and Mathematics
  • Artificial Intelligence
  • Applied Mathematics

Cite this

Improved object categorization and detection using comparative object similarity. / Wang, Gang; Forsyth, David Alexander; Hoiem, Derek W.

In: IEEE transactions on pattern analysis and machine intelligence, Vol. 35, No. 10, 6482138, 03.09.2013, p. 2442-2453.

Research output: Contribution to journalArticle

@article{555ddb068d78485992a804e126215a1f,
title = "Improved object categorization and detection using comparative object similarity",
abstract = "Due to the intrinsic long-tailed distribution of objects in the real world, we are unlikely to be able to train an object recognizer/detector with many visual examples for each category. We have to share visual knowledge between object categories to enable learning with few or no training examples. In this paper, we show that local object similarity information - statements that pairs of categories are similar or dissimilar - is a very useful cue to tie different categories to each other for effective knowledge transfer. The key insight: Given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. To exploit this category-dependent similarity regularization, we develop a regularized kernel machine algorithm to train kernel classifiers for categories with few or no training examples. We also adapt the state-of-the-art object detector to encode object similarity constraints. Our experiments on hundreds of categories from the Labelme dataset show that our regularized kernel classifiers can make significant improvement on object categorization. We also evaluate the improved object detector on the PASCAL VOC 2007 benchmark dataset.",
keywords = "Comparative object similarity, PASCAL VOC, SVM, deformable part model, kernel machines, object categorization, object detection, sharing",
author = "Gang Wang and Forsyth, {David Alexander} and Hoiem, {Derek W}",
year = "2013",
month = "9",
day = "3",
doi = "10.1109/TPAMI.2013.58",
language = "English (US)",
volume = "35",
pages = "2442--2453",
journal = "IEEE Transactions on Pattern Analysis and Machine Intelligence",
issn = "0162-8828",
publisher = "IEEE Computer Society",
number = "10",

}

TY - JOUR

T1 - Improved object categorization and detection using comparative object similarity

AU - Wang, Gang

AU - Forsyth, David Alexander

AU - Hoiem, Derek W

PY - 2013/9/3

Y1 - 2013/9/3

N2 - Due to the intrinsic long-tailed distribution of objects in the real world, we are unlikely to be able to train an object recognizer/detector with many visual examples for each category. We have to share visual knowledge between object categories to enable learning with few or no training examples. In this paper, we show that local object similarity information - statements that pairs of categories are similar or dissimilar - is a very useful cue to tie different categories to each other for effective knowledge transfer. The key insight: Given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. To exploit this category-dependent similarity regularization, we develop a regularized kernel machine algorithm to train kernel classifiers for categories with few or no training examples. We also adapt the state-of-the-art object detector to encode object similarity constraints. Our experiments on hundreds of categories from the Labelme dataset show that our regularized kernel classifiers can make significant improvement on object categorization. We also evaluate the improved object detector on the PASCAL VOC 2007 benchmark dataset.

AB - Due to the intrinsic long-tailed distribution of objects in the real world, we are unlikely to be able to train an object recognizer/detector with many visual examples for each category. We have to share visual knowledge between object categories to enable learning with few or no training examples. In this paper, we show that local object similarity information - statements that pairs of categories are similar or dissimilar - is a very useful cue to tie different categories to each other for effective knowledge transfer. The key insight: Given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. To exploit this category-dependent similarity regularization, we develop a regularized kernel machine algorithm to train kernel classifiers for categories with few or no training examples. We also adapt the state-of-the-art object detector to encode object similarity constraints. Our experiments on hundreds of categories from the Labelme dataset show that our regularized kernel classifiers can make significant improvement on object categorization. We also evaluate the improved object detector on the PASCAL VOC 2007 benchmark dataset.

KW - Comparative object similarity

KW - PASCAL VOC

KW - SVM

KW - deformable part model

KW - kernel machines

KW - object categorization

KW - object detection

KW - sharing

UR - http://www.scopus.com/inward/record.url?scp=84883165952&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84883165952&partnerID=8YFLogxK

U2 - 10.1109/TPAMI.2013.58

DO - 10.1109/TPAMI.2013.58

M3 - Article

C2 - 23969388

AN - SCOPUS:84883165952

VL - 35

SP - 2442

EP - 2453

JO - IEEE Transactions on Pattern Analysis and Machine Intelligence

JF - IEEE Transactions on Pattern Analysis and Machine Intelligence

SN - 0162-8828

IS - 10

M1 - 6482138

ER -