Detecting People in Artwork with CNNs

Nicholas Westlake; Peter Hall; Hongping Cai

doi:10.1007/978-3-319-46604-0_57

Detecting People in Artwork with CNNs

Nicholas Westlake
, Peter Hall
, Hongping Cai

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

65 Citations (Scopus)

471 Downloads (Pure)

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 fine-tuning 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 language	English
Title of host publication	Computer Vision – ECCV 2016 Workshops
Subtitle of host publication	Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, Proceedings, Part I
Editors	Gang Hua, Hervé Jégou
Publisher	Springer Berlin Heidelberg
Pages	825-841
Number of pages	17
ISBN (Electronic)	9783319466040
ISBN (Print)	9783319466033
DOIs	https://doi.org/10.1007/978-3-319-46604-0_57
Publication status	Published - 18 Sept 2016

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	9913
ISSN (Print)	0302-9743

Keywords

CNNs
Cross-depiction problem
Object recognition

Access to Document

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

Full-text PDF (accepted author manuscript)
This is the author accepted manuscript (AAM). The final published version (version of record) is available online via Springer at https://link.springer.com/chapter/10.1007%2F978-3-319-46604-0_57. Please refer to any applicable terms of use of the publisher.
Accepted author manuscript, 6.02 MB

Handle.net

Persistent link

Cite this

@inproceedings{51ae8cca672544ab828e6a22799ce06c,

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 fine-tuning 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{\textquoteright}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 Peter Hall and Hongping Cai",

year = "2016",

month = sep,

day = "18",

doi = "10.1007/978-3-319-46604-0\_57",

language = "English",

isbn = "9783319466033",

series = "Lecture Notes in Computer Science",

publisher = "Springer Berlin Heidelberg",

pages = "825--841",

editor = "Gang Hua and Herv{\'e} J{\'e}gou",

booktitle = "Computer Vision – ECCV 2016 Workshops",

address = "Germany",

}

Detecting People in Artwork with CNNs. / Westlake, Nicholas; Hall, Peter; Cai, Hongping.
Computer Vision – ECCV 2016 Workshops: Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, Proceedings, Part I. ed. / Gang Hua; Hervé Jégou. Springer Berlin Heidelberg, 2016. p. 825-841 (Lecture Notes in Computer Science; Vol. 9913).

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

TY - GEN

T1 - Detecting People in Artwork with CNNs

AU - Westlake, Nicholas

AU - Hall, Peter

AU - Cai, Hongping

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 fine-tuning 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 fine-tuning 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

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

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

M3 - Conference Contribution (Conference Proceeding)

SN - 9783319466033

T3 - Lecture Notes in Computer Science

SP - 825

EP - 841

BT - Computer Vision – ECCV 2016 Workshops

A2 - Hua, Gang

A2 - Jégou, Hervé

PB - Springer Berlin Heidelberg

ER -

Detecting People in Artwork with CNNs

Abstract

Publication series

Keywords

Access to Document

Handle.net

Fingerprint

Cite this