Consistent Semantic Attacks on Optical Flow

Tom Koren; Lior Talker; Michael Dinerstein; Ran Vitek

doi:10.1007/978-3-031-26293-7_30

Consistent Semantic Attacks on Optical Flow

Tom Koren, Lior Talker, Michael Dinerstein, Ran Vitek

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

Abstract

We present a novel approach for semantically targeted adversarial attacks on Optical Flow. In such attacks the goal is to corrupt the flow predictions of a specific object category or instance. Usually, an attacker seeks to hide the adversarial perturbations in the input. However, a quick scan of the output reveals the attack. In contrast, our method helps to hide the attacker’s intent in the output flow as well. We achieve this thanks to a regularization term that encourages off-target consistency. We perform extensive tests on leading optical flow models to demonstrate the benefits of our approach in both white-box and black-box settings. Also, we demonstrate the effectiveness of our attack on subsequent tasks that depend on the optical flow.

Original language	English
Title of host publication	Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings
Editors	Lei Wang, Juergen Gall, Tat-Jun Chin, Imari Sato, Rama Chellappa
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	501-517
Number of pages	17
ISBN (Print)	9783031262920
DOIs	https://doi.org/10.1007/978-3-031-26293-7_30
State	Published - 2023
Externally published	Yes
Event	16th Asian Conference on Computer Vision, ACCV 2022 - Macao, China Duration: 4 Dec 2022 → 8 Dec 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13847 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	16th Asian Conference on Computer Vision, ACCV 2022
Country/Territory	China
City	Macao
Period	4/12/22 → 8/12/22

Bibliographical note

Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Keywords

Adversarial attacks
Optical flow
Semantic attacks

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-031-26293-7_30

Cite this

Koren, T., Talker, L., Dinerstein, M., & Vitek, R. (2023). Consistent Semantic Attacks on Optical Flow. In L. Wang, J. Gall, T.-J. Chin, I. Sato, & R. Chellappa (Eds.), Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings (pp. 501-517). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13847 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-26293-7_30

Koren, Tom ; Talker, Lior ; Dinerstein, Michael et al. / Consistent Semantic Attacks on Optical Flow. Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings. editor / Lei Wang ; Juergen Gall ; Tat-Jun Chin ; Imari Sato ; Rama Chellappa. Springer Science and Business Media Deutschland GmbH, 2023. pp. 501-517 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{8bf8eca193d04a119288449c3a385abc,

title = "Consistent Semantic Attacks on Optical Flow",

abstract = "We present a novel approach for semantically targeted adversarial attacks on Optical Flow. In such attacks the goal is to corrupt the flow predictions of a specific object category or instance. Usually, an attacker seeks to hide the adversarial perturbations in the input. However, a quick scan of the output reveals the attack. In contrast, our method helps to hide the attacker{\textquoteright}s intent in the output flow as well. We achieve this thanks to a regularization term that encourages off-target consistency. We perform extensive tests on leading optical flow models to demonstrate the benefits of our approach in both white-box and black-box settings. Also, we demonstrate the effectiveness of our attack on subsequent tasks that depend on the optical flow.",

keywords = "Adversarial attacks, Optical flow, Semantic attacks",

author = "Tom Koren and Lior Talker and Michael Dinerstein and Ran Vitek",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 16th Asian Conference on Computer Vision, ACCV 2022 ; Conference date: 04-12-2022 Through 08-12-2022",

year = "2023",

doi = "10.1007/978-3-031-26293-7_30",

language = "English",

isbn = "9783031262920",

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 = "501--517",

editor = "Lei Wang and Juergen Gall and Tat-Jun Chin and Imari Sato and Rama Chellappa",

booktitle = "Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings",

address = "Germany",

}

Koren, T, Talker, L, Dinerstein, M & Vitek, R 2023, Consistent Semantic Attacks on Optical Flow. in L Wang, J Gall, T-J Chin, I Sato & R Chellappa (eds), Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13847 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 501-517, 16th Asian Conference on Computer Vision, ACCV 2022, Macao, China, 4/12/22. https://doi.org/10.1007/978-3-031-26293-7_30

Consistent Semantic Attacks on Optical Flow. / Koren, Tom; Talker, Lior; Dinerstein, Michael et al.
Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings. ed. / Lei Wang; Juergen Gall; Tat-Jun Chin; Imari Sato; Rama Chellappa. Springer Science and Business Media Deutschland GmbH, 2023. p. 501-517 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13847 LNCS).

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

TY - GEN

T1 - Consistent Semantic Attacks on Optical Flow

AU - Koren, Tom

AU - Talker, Lior

AU - Dinerstein, Michael

AU - Vitek, Ran

PY - 2023

Y1 - 2023

N2 - We present a novel approach for semantically targeted adversarial attacks on Optical Flow. In such attacks the goal is to corrupt the flow predictions of a specific object category or instance. Usually, an attacker seeks to hide the adversarial perturbations in the input. However, a quick scan of the output reveals the attack. In contrast, our method helps to hide the attacker’s intent in the output flow as well. We achieve this thanks to a regularization term that encourages off-target consistency. We perform extensive tests on leading optical flow models to demonstrate the benefits of our approach in both white-box and black-box settings. Also, we demonstrate the effectiveness of our attack on subsequent tasks that depend on the optical flow.

AB - We present a novel approach for semantically targeted adversarial attacks on Optical Flow. In such attacks the goal is to corrupt the flow predictions of a specific object category or instance. Usually, an attacker seeks to hide the adversarial perturbations in the input. However, a quick scan of the output reveals the attack. In contrast, our method helps to hide the attacker’s intent in the output flow as well. We achieve this thanks to a regularization term that encourages off-target consistency. We perform extensive tests on leading optical flow models to demonstrate the benefits of our approach in both white-box and black-box settings. Also, we demonstrate the effectiveness of our attack on subsequent tasks that depend on the optical flow.

KW - Adversarial attacks

KW - Optical flow

KW - Semantic attacks

UR - http://www.scopus.com/inward/record.url?scp=85151049362&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-26293-7_30

DO - 10.1007/978-3-031-26293-7_30

M3 - Conference contribution

AN - SCOPUS:85151049362

SN - 9783031262920

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

SP - 501

EP - 517

BT - Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings

A2 - Wang, Lei

A2 - Gall, Juergen

A2 - Chin, Tat-Jun

A2 - Sato, Imari

A2 - Chellappa, Rama

PB - Springer Science and Business Media Deutschland GmbH

T2 - 16th Asian Conference on Computer Vision, ACCV 2022

Y2 - 4 December 2022 through 8 December 2022

ER -

Koren T, Talker L, Dinerstein M, Vitek R. Consistent Semantic Attacks on Optical Flow. In Wang L, Gall J, Chin TJ, Sato I, Chellappa R, editors, Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 501-517. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-26293-7_30

Consistent Semantic Attacks on Optical Flow

Abstract

Publication series

Conference

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this