Semantic label sharing for learning with many categories

Robert Fergus, Hector Bernal, Yair Weiss, Antonio Torralba

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In an object recognition scenario with tens of thousands of categories, even a small number of labels per category leads to a very large number of total labels required. We propose a simple method of label sharing between semantically similar categories. We leverage the WordNet hierarchy to define semantic distance between any two categories and use this semantic distance to share labels. Our approach can be used with any classifier. Experimental results on a range of datasets, upto 80 million images and 75,000 categories in size, show that despite the simplicity of the approach, it leads to significant improvements in performance.

Original languageEnglish (US)
Title of host publicationComputer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings
Pages762-775
Number of pages14
Volume6311 LNCS
EditionPART 1
DOIs
StatePublished - 2010
Event11th European Conference on Computer Vision, ECCV 2010 - Heraklion, Crete, Greece
Duration: Sep 5 2010Sep 11 2010

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 1
Volume6311 LNCS
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other11th European Conference on Computer Vision, ECCV 2010
CountryGreece
CityHeraklion, Crete
Period9/5/109/11/10

Fingerprint

Labels
Sharing
Semantics
Object recognition
WordNet
Object Recognition
Classifiers
Leverage
Simplicity
Classifier
Learning
Scenarios
Experimental Results
Range of data

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Fergus, R., Bernal, H., Weiss, Y., & Torralba, A. (2010). Semantic label sharing for learning with many categories. In Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings (PART 1 ed., Vol. 6311 LNCS, pp. 762-775). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6311 LNCS, No. PART 1). https://doi.org/10.1007/978-3-642-15549-9_55

Semantic label sharing for learning with many categories. / Fergus, Robert; Bernal, Hector; Weiss, Yair; Torralba, Antonio.

Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings. Vol. 6311 LNCS PART 1. ed. 2010. p. 762-775 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6311 LNCS, No. PART 1).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Fergus, R, Bernal, H, Weiss, Y & Torralba, A 2010, Semantic label sharing for learning with many categories. in Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings. PART 1 edn, vol. 6311 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 1, vol. 6311 LNCS, pp. 762-775, 11th European Conference on Computer Vision, ECCV 2010, Heraklion, Crete, Greece, 9/5/10. https://doi.org/10.1007/978-3-642-15549-9_55
Fergus R, Bernal H, Weiss Y, Torralba A. Semantic label sharing for learning with many categories. In Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings. PART 1 ed. Vol. 6311 LNCS. 2010. p. 762-775. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1). https://doi.org/10.1007/978-3-642-15549-9_55
Fergus, Robert ; Bernal, Hector ; Weiss, Yair ; Torralba, Antonio. / Semantic label sharing for learning with many categories. Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings. Vol. 6311 LNCS PART 1. ed. 2010. pp. 762-775 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1).
@inproceedings{bc77a7b7f7dd410894fd7a0b99996304,
title = "Semantic label sharing for learning with many categories",
abstract = "In an object recognition scenario with tens of thousands of categories, even a small number of labels per category leads to a very large number of total labels required. We propose a simple method of label sharing between semantically similar categories. We leverage the WordNet hierarchy to define semantic distance between any two categories and use this semantic distance to share labels. Our approach can be used with any classifier. Experimental results on a range of datasets, upto 80 million images and 75,000 categories in size, show that despite the simplicity of the approach, it leads to significant improvements in performance.",
author = "Robert Fergus and Hector Bernal and Yair Weiss and Antonio Torralba",
year = "2010",
doi = "10.1007/978-3-642-15549-9_55",
language = "English (US)",
isbn = "3642155480",
volume = "6311 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
number = "PART 1",
pages = "762--775",
booktitle = "Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings",
edition = "PART 1",

}

TY - GEN

T1 - Semantic label sharing for learning with many categories

AU - Fergus, Robert

AU - Bernal, Hector

AU - Weiss, Yair

AU - Torralba, Antonio

PY - 2010

Y1 - 2010

N2 - In an object recognition scenario with tens of thousands of categories, even a small number of labels per category leads to a very large number of total labels required. We propose a simple method of label sharing between semantically similar categories. We leverage the WordNet hierarchy to define semantic distance between any two categories and use this semantic distance to share labels. Our approach can be used with any classifier. Experimental results on a range of datasets, upto 80 million images and 75,000 categories in size, show that despite the simplicity of the approach, it leads to significant improvements in performance.

AB - In an object recognition scenario with tens of thousands of categories, even a small number of labels per category leads to a very large number of total labels required. We propose a simple method of label sharing between semantically similar categories. We leverage the WordNet hierarchy to define semantic distance between any two categories and use this semantic distance to share labels. Our approach can be used with any classifier. Experimental results on a range of datasets, upto 80 million images and 75,000 categories in size, show that despite the simplicity of the approach, it leads to significant improvements in performance.

UR - http://www.scopus.com/inward/record.url?scp=78149329142&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=78149329142&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-15549-9_55

DO - 10.1007/978-3-642-15549-9_55

M3 - Conference contribution

AN - SCOPUS:78149329142

SN - 3642155480

SN - 9783642155482

VL - 6311 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 762

EP - 775

BT - Computer Vision, ECCV 2010 - 11th European Conference on Computer Vision, Proceedings

ER -