Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization

Jun Ho Choi; Huan Zhang; Jun Hyuk Kim; Cho Jui Hsieh; Jong Seok Lee

doi:10.1007/978-3-030-69538-5_19

Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization

Jun Ho Choi, Huan Zhang, Jun Hyuk Kim, Cho Jui Hsieh, Jong Seok Lee

School of Integrated Technology

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

Abstract

Image super-resolution has been widely employed in various applications with boosted performance thanks to the deep learning techniques. However, many deep learning-based models are highly vulnerable to adversarial attacks, which is also applied to super-resolution models in recent studies. In this paper, we propose a defense method that is formulated as an entropy regularization loss for model training, which can be augmented to the original training loss of super-resolution models. We show that various state-of-the-art super-resolution models trained with our defense method are more robust against adversarial attacks than their original versions. To the best of our knowledge, this is the first attempt of adversarial defense for deep super-resolution models.

Original language	English
Title of host publication	Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers
Editors	Hiroshi Ishikawa, Cheng-Lin Liu, Tomas Pajdla, Jianbo Shi
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	301-317
Number of pages	17
ISBN (Print)	9783030695378
DOIs	https://doi.org/10.1007/978-3-030-69538-5_19
Publication status	Published - 2021
Event	15th Asian Conference on Computer Vision, ACCV 2020 - Virtual, Online Duration: 2020 Nov 30 → 2020 Dec 4

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12625 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	15th Asian Conference on Computer Vision, ACCV 2020
City	Virtual, Online
Period	20/11/30 → 20/12/4

Bibliographical note

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

All Science Journal Classification (ASJC) codes

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-030-69538-5_19

Cite this

Choi, J. H., Zhang, H., Kim, J. H., Hsieh, C. J., & Lee, J. S. (2021). Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization. In H. Ishikawa, C.-L. Liu, T. Pajdla, & J. Shi (Eds.), Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers (pp. 301-317). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12625 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-69538-5_19

Choi, Jun Ho ; Zhang, Huan ; Kim, Jun Hyuk et al. / Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization. Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers. editor / Hiroshi Ishikawa ; Cheng-Lin Liu ; Tomas Pajdla ; Jianbo Shi. Springer Science and Business Media Deutschland GmbH, 2021. pp. 301-317 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{06260c4f5a0b47c895134b0cead38af2,

title = "Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization",

abstract = "Image super-resolution has been widely employed in various applications with boosted performance thanks to the deep learning techniques. However, many deep learning-based models are highly vulnerable to adversarial attacks, which is also applied to super-resolution models in recent studies. In this paper, we propose a defense method that is formulated as an entropy regularization loss for model training, which can be augmented to the original training loss of super-resolution models. We show that various state-of-the-art super-resolution models trained with our defense method are more robust against adversarial attacks than their original versions. To the best of our knowledge, this is the first attempt of adversarial defense for deep super-resolution models.",

author = "Choi, {Jun Ho} and Huan Zhang and Kim, {Jun Hyuk} and Hsieh, {Cho Jui} and Lee, {Jong Seok}",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 15th Asian Conference on Computer Vision, ACCV 2020 ; Conference date: 30-11-2020 Through 04-12-2020",

year = "2021",

doi = "10.1007/978-3-030-69538-5_19",

language = "English",

isbn = "9783030695378",

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

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "301--317",

editor = "Hiroshi Ishikawa and Cheng-Lin Liu and Tomas Pajdla and Jianbo Shi",

booktitle = "Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers",

address = "Germany",

}

Choi, JH, Zhang, H, Kim, JH, Hsieh, CJ & Lee, JS 2021, Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization. in H Ishikawa, C-L Liu, T Pajdla & J Shi (eds), Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12625 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 301-317, 15th Asian Conference on Computer Vision, ACCV 2020, Virtual, Online, 20/11/30. https://doi.org/10.1007/978-3-030-69538-5_19

Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization. / Choi, Jun Ho; Zhang, Huan; Kim, Jun Hyuk et al.
Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers. ed. / Hiroshi Ishikawa; Cheng-Lin Liu; Tomas Pajdla; Jianbo Shi. Springer Science and Business Media Deutschland GmbH, 2021. p. 301-317 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12625 LNCS).

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

TY - GEN

T1 - Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization

AU - Choi, Jun Ho

AU - Zhang, Huan

AU - Kim, Jun Hyuk

AU - Hsieh, Cho Jui

AU - Lee, Jong Seok

PY - 2021

Y1 - 2021

N2 - Image super-resolution has been widely employed in various applications with boosted performance thanks to the deep learning techniques. However, many deep learning-based models are highly vulnerable to adversarial attacks, which is also applied to super-resolution models in recent studies. In this paper, we propose a defense method that is formulated as an entropy regularization loss for model training, which can be augmented to the original training loss of super-resolution models. We show that various state-of-the-art super-resolution models trained with our defense method are more robust against adversarial attacks than their original versions. To the best of our knowledge, this is the first attempt of adversarial defense for deep super-resolution models.

AB - Image super-resolution has been widely employed in various applications with boosted performance thanks to the deep learning techniques. However, many deep learning-based models are highly vulnerable to adversarial attacks, which is also applied to super-resolution models in recent studies. In this paper, we propose a defense method that is formulated as an entropy regularization loss for model training, which can be augmented to the original training loss of super-resolution models. We show that various state-of-the-art super-resolution models trained with our defense method are more robust against adversarial attacks than their original versions. To the best of our knowledge, this is the first attempt of adversarial defense for deep super-resolution models.

UR - http://www.scopus.com/inward/record.url?scp=85103279124&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85103279124&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-69538-5_19

DO - 10.1007/978-3-030-69538-5_19

M3 - Conference contribution

AN - SCOPUS:85103279124

SN - 9783030695378

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

SP - 301

EP - 317

BT - Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers

A2 - Ishikawa, Hiroshi

A2 - Liu, Cheng-Lin

A2 - Pajdla, Tomas

A2 - Shi, Jianbo

PB - Springer Science and Business Media Deutschland GmbH

T2 - 15th Asian Conference on Computer Vision, ACCV 2020

Y2 - 30 November 2020 through 4 December 2020

ER -

Choi JH, Zhang H, Kim JH, Hsieh CJ, Lee JS. Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization. In Ishikawa H, Liu CL, Pajdla T, Shi J, editors, Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers. Springer Science and Business Media Deutschland GmbH. 2021. p. 301-317. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-69538-5_19

Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization

Abstract

Publication series

Conference

Bibliographical note

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this