The devil of face recognition is in the noise

Fei Wang; Liren Chen; Cheng Li; Shiyao Huang; Yanjie Chen; Chen Qian; Chen Change Loy

doi:10.1007/978-3-030-01240-3_47

The devil of face recognition is in the noise

Fei Wang^*, Liren Chen, Cheng Li, Shiyao Huang, Yanjie Chen, Chen Qian, Chen Change Loy

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

33 Citations (Scopus)

Abstract

The growing scale of face recognition datasets empowers us to train strong convolutional networks for face recognition. While a variety of architectures and loss functions have been devised, we still have a limited understanding of the source and consequence of label noise inherent in existing datasets. We make the following contributions: (1) We contribute cleaned subsets of popular face databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new large-scale noise-controlled IMDb-Face dataset. (2) With the original datasets and cleaned subsets, we profile and analyze label noise properties of MegaFace and MS-Celeb-1M. We show that a few orders more samples are needed to achieve the same accuracy yielded by a clean subset. (3) We study the association between different types of noise, i.e., label flips and outliers, with the accuracy of face recognition models. (4) We investigate ways to improve data cleanliness, including a comprehensive user study on the influence of data labeling strategies to annotation accuracy. The IMDb-Face dataset has been released on https://github.com/fwang91/IMDb-Face.

Original language	English
Title of host publication	Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
Editors	Martial Hebert, Vittorio Ferrari, Cristian Sminchisescu, Yair Weiss
Publisher	Springer Verlag
Pages	780-795
Number of pages	16
ISBN (Print)	9783030012397
DOIs	https://doi.org/10.1007/978-3-030-01240-3_47
Publication status	Published - 2018
Externally published	Yes
Event	15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany Duration: Sept 8 2018 → Sept 14 2018

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11213 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	15th European Conference on Computer Vision, ECCV 2018
Country/Territory	Germany
City	Munich
Period	9/8/18 → 9/14/18

Bibliographical note

Publisher Copyright:
© Springer Nature Switzerland AG 2018.

ASJC Scopus Subject Areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-030-01240-3_47

Cite this

Wang, F., Chen, L., Li, C., Huang, S., Chen, Y., Qian, C., & Loy, C. C. (2018). The devil of face recognition is in the noise. In M. Hebert, V. Ferrari, C. Sminchisescu, & Y. Weiss (Eds.), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings (pp. 780-795). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11213 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-01240-3_47

Wang, Fei ; Chen, Liren ; Li, Cheng et al. / The devil of face recognition is in the noise. Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. editor / Martial Hebert ; Vittorio Ferrari ; Cristian Sminchisescu ; Yair Weiss. Springer Verlag, 2018. pp. 780-795 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{562c96eef8534d3fab8fbfecdb09dfd1,

title = "The devil of face recognition is in the noise",

abstract = "The growing scale of face recognition datasets empowers us to train strong convolutional networks for face recognition. While a variety of architectures and loss functions have been devised, we still have a limited understanding of the source and consequence of label noise inherent in existing datasets. We make the following contributions: (1) We contribute cleaned subsets of popular face databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new large-scale noise-controlled IMDb-Face dataset. (2) With the original datasets and cleaned subsets, we profile and analyze label noise properties of MegaFace and MS-Celeb-1M. We show that a few orders more samples are needed to achieve the same accuracy yielded by a clean subset. (3) We study the association between different types of noise, i.e., label flips and outliers, with the accuracy of face recognition models. (4) We investigate ways to improve data cleanliness, including a comprehensive user study on the influence of data labeling strategies to annotation accuracy. The IMDb-Face dataset has been released on https://github.com/fwang91/IMDb-Face.",

author = "Fei Wang and Liren Chen and Cheng Li and Shiyao Huang and Yanjie Chen and Chen Qian and Loy, \{Chen Change\}",

note = "Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2018.; 15th European Conference on Computer Vision, ECCV 2018 ; Conference date: 08-09-2018 Through 14-09-2018",

year = "2018",

doi = "10.1007/978-3-030-01240-3\_47",

language = "English",

isbn = "9783030012397",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "780--795",

editor = "Martial Hebert and Vittorio Ferrari and Cristian Sminchisescu and Yair Weiss",

booktitle = "Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings",

address = "Germany",

}

Wang, F, Chen, L, Li, C, Huang, S, Chen, Y, Qian, C & Loy, CC 2018, The devil of face recognition is in the noise. in M Hebert, V Ferrari, C Sminchisescu & Y Weiss (eds), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11213 LNCS, Springer Verlag, pp. 780-795, 15th European Conference on Computer Vision, ECCV 2018, Munich, Germany, 9/8/18. https://doi.org/10.1007/978-3-030-01240-3_47

The devil of face recognition is in the noise. / Wang, Fei; Chen, Liren; Li, Cheng et al.
Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. ed. / Martial Hebert; Vittorio Ferrari; Cristian Sminchisescu; Yair Weiss. Springer Verlag, 2018. p. 780-795 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11213 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - The devil of face recognition is in the noise

AU - Wang, Fei

AU - Chen, Liren

AU - Li, Cheng

AU - Huang, Shiyao

AU - Chen, Yanjie

AU - Qian, Chen

AU - Loy, Chen Change

N1 - Publisher Copyright: © Springer Nature Switzerland AG 2018.

PY - 2018

Y1 - 2018

N2 - The growing scale of face recognition datasets empowers us to train strong convolutional networks for face recognition. While a variety of architectures and loss functions have been devised, we still have a limited understanding of the source and consequence of label noise inherent in existing datasets. We make the following contributions: (1) We contribute cleaned subsets of popular face databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new large-scale noise-controlled IMDb-Face dataset. (2) With the original datasets and cleaned subsets, we profile and analyze label noise properties of MegaFace and MS-Celeb-1M. We show that a few orders more samples are needed to achieve the same accuracy yielded by a clean subset. (3) We study the association between different types of noise, i.e., label flips and outliers, with the accuracy of face recognition models. (4) We investigate ways to improve data cleanliness, including a comprehensive user study on the influence of data labeling strategies to annotation accuracy. The IMDb-Face dataset has been released on https://github.com/fwang91/IMDb-Face.

AB - The growing scale of face recognition datasets empowers us to train strong convolutional networks for face recognition. While a variety of architectures and loss functions have been devised, we still have a limited understanding of the source and consequence of label noise inherent in existing datasets. We make the following contributions: (1) We contribute cleaned subsets of popular face databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new large-scale noise-controlled IMDb-Face dataset. (2) With the original datasets and cleaned subsets, we profile and analyze label noise properties of MegaFace and MS-Celeb-1M. We show that a few orders more samples are needed to achieve the same accuracy yielded by a clean subset. (3) We study the association between different types of noise, i.e., label flips and outliers, with the accuracy of face recognition models. (4) We investigate ways to improve data cleanliness, including a comprehensive user study on the influence of data labeling strategies to annotation accuracy. The IMDb-Face dataset has been released on https://github.com/fwang91/IMDb-Face.

UR - http://www.scopus.com/inward/record.url?scp=85055112107&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85055112107&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-01240-3_47

DO - 10.1007/978-3-030-01240-3_47

M3 - Conference contribution

AN - SCOPUS:85055112107

SN - 9783030012397

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

SP - 780

EP - 795

BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings

A2 - Hebert, Martial

A2 - Ferrari, Vittorio

A2 - Sminchisescu, Cristian

A2 - Weiss, Yair

PB - Springer Verlag

T2 - 15th European Conference on Computer Vision, ECCV 2018

Y2 - 8 September 2018 through 14 September 2018

ER -

Wang F, Chen L, Li C, Huang S, Chen Y, Qian C et al. The devil of face recognition is in the noise. In Hebert M, Ferrari V, Sminchisescu C, Weiss Y, editors, Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer Verlag. 2018. p. 780-795. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-01240-3_47

The devil of face recognition is in the noise

Abstract

Publication series

Conference

Bibliographical note

ASJC Scopus Subject Areas

Access to Document

Other files and links

Cite this