Out-of-Distribution Detection Without Class Labels

Niv Cohen; Ron Abutbul; Yedid Hoshen

doi:10.1007/978-3-031-25063-7_7

Out-of-Distribution Detection Without Class Labels

Niv Cohen^*, Ron Abutbul, Yedid Hoshen

^*Corresponding author for this work

The Rachel and Selim Benin School of Engineering and Computer Science

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

Abstract

Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consist of multiple semantic classes (e.g. multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover “pseudo-class” labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.

Original language	American English
Title of host publication	Computer Vision – ECCV 2022 Workshops, Proceedings
Editors	Leonid Karlinsky, Tomer Michaeli, Ko Nishino
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	101-117
Number of pages	17
ISBN (Print)	9783031250620
DOIs	https://doi.org/10.1007/978-3-031-25063-7_7
State	Published - 2023
Event	17th European Conference on Computer Vision, ECCV 2022 - Tel Aviv, Israel Duration: 23 Oct 2022 → 27 Oct 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13802 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	17th European Conference on Computer Vision, ECCV 2022
Country/Territory	Israel
City	Tel Aviv
Period	23/10/22 → 27/10/22

Bibliographical note

Funding Information:
Acknowledgements. This work was partly supported by the Malvina and Solomon Pollack scholarship and, the Federmann Cyber Security Research Center in conjunction with the Israel National Cyber Directorate.

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

Access to Document

10.1007/978-3-031-25063-7_7

Cite this

Cohen, N., Abutbul, R., & Hoshen, Y. (2023). Out-of-Distribution Detection Without Class Labels. In L. Karlinsky, T. Michaeli, & K. Nishino (Eds.), Computer Vision – ECCV 2022 Workshops, Proceedings (pp. 101-117). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13802 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-25063-7_7

Cohen, Niv ; Abutbul, Ron ; Hoshen, Yedid. / Out-of-Distribution Detection Without Class Labels. Computer Vision – ECCV 2022 Workshops, Proceedings. editor / Leonid Karlinsky ; Tomer Michaeli ; Ko Nishino. Springer Science and Business Media Deutschland GmbH, 2023. pp. 101-117 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{f9d2ca7f575847bb90ed98a73bba04bd,

title = "Out-of-Distribution Detection Without Class Labels",

abstract = "Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consist of multiple semantic classes (e.g. multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover “pseudo-class” labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.",

author = "Niv Cohen and Ron Abutbul and Yedid Hoshen",

note = "Funding Information: Acknowledgements. This work was partly supported by the Malvina and Solomon Pollack scholarship and, the Federmann Cyber Security Research Center in conjunction with the Israel National Cyber Directorate. Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 17th European Conference on Computer Vision, ECCV 2022 ; Conference date: 23-10-2022 Through 27-10-2022",

year = "2023",

doi = "10.1007/978-3-031-25063-7_7",

language = "American English",

isbn = "9783031250620",

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 = "101--117",

editor = "Leonid Karlinsky and Tomer Michaeli and Ko Nishino",

booktitle = "Computer Vision – ECCV 2022 Workshops, Proceedings",

address = "Germany",

}

Cohen, N, Abutbul, R & Hoshen, Y 2023, Out-of-Distribution Detection Without Class Labels. in L Karlinsky, T Michaeli & K Nishino (eds), Computer Vision – ECCV 2022 Workshops, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13802 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 101-117, 17th European Conference on Computer Vision, ECCV 2022, Tel Aviv, Israel, 23/10/22. https://doi.org/10.1007/978-3-031-25063-7_7

Out-of-Distribution Detection Without Class Labels. / Cohen, Niv; Abutbul, Ron; Hoshen, Yedid.
Computer Vision – ECCV 2022 Workshops, Proceedings. ed. / Leonid Karlinsky; Tomer Michaeli; Ko Nishino. Springer Science and Business Media Deutschland GmbH, 2023. p. 101-117 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13802 LNCS).

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

TY - GEN

T1 - Out-of-Distribution Detection Without Class Labels

AU - Cohen, Niv

AU - Abutbul, Ron

AU - Hoshen, Yedid

N1 - Funding Information: Acknowledgements. This work was partly supported by the Malvina and Solomon Pollack scholarship and, the Federmann Cyber Security Research Center in conjunction with the Israel National Cyber Directorate. Publisher Copyright: © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

PY - 2023

Y1 - 2023

N2 - Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consist of multiple semantic classes (e.g. multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover “pseudo-class” labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.

AB - Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consist of multiple semantic classes (e.g. multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover “pseudo-class” labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.

UR - http://www.scopus.com/inward/record.url?scp=85150589836&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-25063-7_7

DO - 10.1007/978-3-031-25063-7_7

M3 - Conference contribution

AN - SCOPUS:85150589836

SN - 9783031250620

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

SP - 101

EP - 117

BT - Computer Vision – ECCV 2022 Workshops, Proceedings

A2 - Karlinsky, Leonid

A2 - Michaeli, Tomer

A2 - Nishino, Ko

PB - Springer Science and Business Media Deutschland GmbH

T2 - 17th European Conference on Computer Vision, ECCV 2022

Y2 - 23 October 2022 through 27 October 2022

ER -

Cohen N, Abutbul R, Hoshen Y. Out-of-Distribution Detection Without Class Labels. In Karlinsky L, Michaeli T, Nishino K, editors, Computer Vision – ECCV 2022 Workshops, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 101-117. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-25063-7_7

Out-of-Distribution Detection Without Class Labels

Abstract

Publication series

Conference

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this