Detecting people in artwork with CNNs

Nicholas Westlake, Hongping Cai, Peter Hall

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

5 Citations (Scopus)

Abstract

CNNs have massively improved performance in object detection in photographs. However research into object detection in artwork remains limited. We show state-of-the-art performance on a challenging dataset, People-Art, which contains people from photos, cartoons and 41 different artwork movements. We achieve this high performance by finetuning a CNN for this task, thus also demonstrating that training CNNs on photos results in overfitting for photos: only the first three or four layers transfer from photos to artwork. Although the CNN’s performance is the highest yet, it remains less than 60% AP, suggesting further work is needed for the cross-depiction problem.

Original languageEnglish
Title of host publicationComputer Vision - ECCV 2016 Workshops. ECCV 2016.
EditorsGang Hua, Hervé Jégou
PublisherSpringer Verlag
Pages825-841
Number of pages17
ISBN (Print)9783319466033
DOIs
Publication statusPublished - 18 Sep 2016
Event14th European Conference on Computer Vision, ECCV 2016 - Amsterdam, Netherlands
Duration: 8 Oct 201616 Oct 2016

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9913 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference14th European Conference on Computer Vision, ECCV 2016
CountryNetherlands
CityAmsterdam
Period8/10/1616/10/16

Keywords

  • CNNs
  • Cross-depiction problem
  • Object recognition

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Westlake, N., Cai, H., & Hall, P. (2016). Detecting people in artwork with CNNs. In G. Hua, & H. Jégou (Eds.), Computer Vision - ECCV 2016 Workshops. ECCV 2016. (pp. 825-841). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9913 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-46604-0_57

Detecting people in artwork with CNNs. / Westlake, Nicholas; Cai, Hongping; Hall, Peter.

Computer Vision - ECCV 2016 Workshops. ECCV 2016.. ed. / Gang Hua; Hervé Jégou. Springer Verlag, 2016. p. 825-841 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9913 LNCS).

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

Westlake, N, Cai, H & Hall, P 2016, Detecting people in artwork with CNNs. in G Hua & H Jégou (eds), Computer Vision - ECCV 2016 Workshops. ECCV 2016.. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9913 LNCS, Springer Verlag, pp. 825-841, 14th European Conference on Computer Vision, ECCV 2016, Amsterdam, Netherlands, 8/10/16. https://doi.org/10.1007/978-3-319-46604-0_57
Westlake N, Cai H, Hall P. Detecting people in artwork with CNNs. In Hua G, Jégou H, editors, Computer Vision - ECCV 2016 Workshops. ECCV 2016.. Springer Verlag. 2016. p. 825-841. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-46604-0_57
Westlake, Nicholas ; Cai, Hongping ; Hall, Peter. / Detecting people in artwork with CNNs. Computer Vision - ECCV 2016 Workshops. ECCV 2016.. editor / Gang Hua ; Hervé Jégou. Springer Verlag, 2016. pp. 825-841 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{c50337d1210a48ca8ad9bd3ce29e6ce4,
title = "Detecting people in artwork with CNNs",
abstract = "CNNs have massively improved performance in object detection in photographs. However research into object detection in artwork remains limited. We show state-of-the-art performance on a challenging dataset, People-Art, which contains people from photos, cartoons and 41 different artwork movements. We achieve this high performance by finetuning a CNN for this task, thus also demonstrating that training CNNs on photos results in overfitting for photos: only the first three or four layers transfer from photos to artwork. Although the CNN’s performance is the highest yet, it remains less than 60{\%} AP, suggesting further work is needed for the cross-depiction problem.",
keywords = "CNNs, Cross-depiction problem, Object recognition",
author = "Nicholas Westlake and Hongping Cai and Peter Hall",
year = "2016",
month = "9",
day = "18",
doi = "10.1007/978-3-319-46604-0_57",
language = "English",
isbn = "9783319466033",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "825--841",
editor = "Gang Hua and Herv{\'e} J{\'e}gou",
booktitle = "Computer Vision - ECCV 2016 Workshops. ECCV 2016.",

}

TY - GEN

T1 - Detecting people in artwork with CNNs

AU - Westlake, Nicholas

AU - Cai, Hongping

AU - Hall, Peter

PY - 2016/9/18

Y1 - 2016/9/18

N2 - CNNs have massively improved performance in object detection in photographs. However research into object detection in artwork remains limited. We show state-of-the-art performance on a challenging dataset, People-Art, which contains people from photos, cartoons and 41 different artwork movements. We achieve this high performance by finetuning a CNN for this task, thus also demonstrating that training CNNs on photos results in overfitting for photos: only the first three or four layers transfer from photos to artwork. Although the CNN’s performance is the highest yet, it remains less than 60% AP, suggesting further work is needed for the cross-depiction problem.

AB - CNNs have massively improved performance in object detection in photographs. However research into object detection in artwork remains limited. We show state-of-the-art performance on a challenging dataset, People-Art, which contains people from photos, cartoons and 41 different artwork movements. We achieve this high performance by finetuning a CNN for this task, thus also demonstrating that training CNNs on photos results in overfitting for photos: only the first three or four layers transfer from photos to artwork. Although the CNN’s performance is the highest yet, it remains less than 60% AP, suggesting further work is needed for the cross-depiction problem.

KW - CNNs

KW - Cross-depiction problem

KW - Object recognition

UR - http://www.scopus.com/inward/record.url?scp=84989834205&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-46604-0_57

DO - 10.1007/978-3-319-46604-0_57

M3 - Conference contribution

SN - 9783319466033

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

SP - 825

EP - 841

BT - Computer Vision - ECCV 2016 Workshops. ECCV 2016.

A2 - Hua, Gang

A2 - Jégou, Hervé

PB - Springer Verlag

ER -