Out-of-Distribution Detection for Adaptive Computer Vision

Simon Kristoffersson Lind; Rudolph Triebel; Luigi Nardi; Volker Krueger

doi:10.1007/978-3-031-31438-4_21

Out-of-Distribution Detection for Adaptive Computer Vision

Simon Kristoffersson Lind, Rudolph Triebel, Luigi Nardi, Volker Krueger

Computation, Information and Technology

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

Abstract

It is well known that computer vision can be unreliable when faced with previously unseen imaging conditions. This paper proposes a method to adapt camera parameters according to a normalizing flow-based out-of-distibution detector. A small-scale study is conducted which shows that adapting camera parameters according to this out-of-distibution detector leads to an average increase of 3 to 4% points in mAP, mAR and F1 performance metrics of a YOLOv4 object detector. As a secondary result, this paper also shows that it is possible to train a normalizing flow model for out-of-distribution detection on the COCO dataset, which is larger and more diverse than most benchmarks for out-of-distibution detectors.

Original language	English
Title of host publication	Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings
Editors	Rikke Gade, Michael Felsberg, Joni-Kristian Kämäräinen
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	311-325
Number of pages	15
ISBN (Print)	9783031314377
DOIs	https://doi.org/10.1007/978-3-031-31438-4_21
State	Published - 2023
Event	23nd Scandinavian Conference on Image Analysis, SCIA 2023 - Lapland, Finland Duration: 18 Apr 2023 → 21 Apr 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13886 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	23nd Scandinavian Conference on Image Analysis, SCIA 2023
Country/Territory	Finland
City	Lapland
Period	18/04/23 → 21/04/23

Keywords

Autonomous Systems
Normalizing Flows
Object Detection
Out-of-Distribution Detection

Access to Document

10.1007/978-3-031-31438-4_21

Cite this

Kristoffersson Lind, S., Triebel, R., Nardi, L., & Krueger, V. (2023). Out-of-Distribution Detection for Adaptive Computer Vision. In R. Gade, M. Felsberg, & J.-K. Kämäräinen (Eds.), Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings (pp. 311-325). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13886 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-31438-4_21

Kristoffersson Lind, Simon ; Triebel, Rudolph ; Nardi, Luigi et al. / Out-of-Distribution Detection for Adaptive Computer Vision. Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings. editor / Rikke Gade ; Michael Felsberg ; Joni-Kristian Kämäräinen. Springer Science and Business Media Deutschland GmbH, 2023. pp. 311-325 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{ad6c61b5341543c38509351fbca7340a,

title = "Out-of-Distribution Detection for Adaptive Computer Vision",

abstract = "It is well known that computer vision can be unreliable when faced with previously unseen imaging conditions. This paper proposes a method to adapt camera parameters according to a normalizing flow-based out-of-distibution detector. A small-scale study is conducted which shows that adapting camera parameters according to this out-of-distibution detector leads to an average increase of 3 to 4% points in mAP, mAR and F1 performance metrics of a YOLOv4 object detector. As a secondary result, this paper also shows that it is possible to train a normalizing flow model for out-of-distribution detection on the COCO dataset, which is larger and more diverse than most benchmarks for out-of-distibution detectors.",

keywords = "Autonomous Systems, Normalizing Flows, Object Detection, Out-of-Distribution Detection",

author = "{Kristoffersson Lind}, Simon and Rudolph Triebel and Luigi Nardi and Volker Krueger",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 23nd Scandinavian Conference on Image Analysis, SCIA 2023 ; Conference date: 18-04-2023 Through 21-04-2023",

year = "2023",

doi = "10.1007/978-3-031-31438-4_21",

language = "English",

isbn = "9783031314377",

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 = "311--325",

editor = "Rikke Gade and Michael Felsberg and Joni-Kristian K{\"a}m{\"a}r{\"a}inen",

booktitle = "Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings",

}

Kristoffersson Lind, S, Triebel, R, Nardi, L & Krueger, V 2023, Out-of-Distribution Detection for Adaptive Computer Vision. in R Gade, M Felsberg & J-K Kämäräinen (eds), Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13886 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 311-325, 23nd Scandinavian Conference on Image Analysis, SCIA 2023, Lapland, Finland, 18/04/23. https://doi.org/10.1007/978-3-031-31438-4_21

Out-of-Distribution Detection for Adaptive Computer Vision. / Kristoffersson Lind, Simon; Triebel, Rudolph; Nardi, Luigi et al.
Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings. ed. / Rikke Gade; Michael Felsberg; Joni-Kristian Kämäräinen. Springer Science and Business Media Deutschland GmbH, 2023. p. 311-325 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13886 LNCS).

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

TY - GEN

T1 - Out-of-Distribution Detection for Adaptive Computer Vision

AU - Kristoffersson Lind, Simon

AU - Triebel, Rudolph

AU - Nardi, Luigi

AU - Krueger, Volker

PY - 2023

Y1 - 2023

N2 - It is well known that computer vision can be unreliable when faced with previously unseen imaging conditions. This paper proposes a method to adapt camera parameters according to a normalizing flow-based out-of-distibution detector. A small-scale study is conducted which shows that adapting camera parameters according to this out-of-distibution detector leads to an average increase of 3 to 4% points in mAP, mAR and F1 performance metrics of a YOLOv4 object detector. As a secondary result, this paper also shows that it is possible to train a normalizing flow model for out-of-distribution detection on the COCO dataset, which is larger and more diverse than most benchmarks for out-of-distibution detectors.

AB - It is well known that computer vision can be unreliable when faced with previously unseen imaging conditions. This paper proposes a method to adapt camera parameters according to a normalizing flow-based out-of-distibution detector. A small-scale study is conducted which shows that adapting camera parameters according to this out-of-distibution detector leads to an average increase of 3 to 4% points in mAP, mAR and F1 performance metrics of a YOLOv4 object detector. As a secondary result, this paper also shows that it is possible to train a normalizing flow model for out-of-distribution detection on the COCO dataset, which is larger and more diverse than most benchmarks for out-of-distibution detectors.

KW - Autonomous Systems

KW - Normalizing Flows

KW - Object Detection

KW - Out-of-Distribution Detection

UR - http://www.scopus.com/inward/record.url?scp=85161436449&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-31438-4_21

DO - 10.1007/978-3-031-31438-4_21

M3 - Conference contribution

AN - SCOPUS:85161436449

SN - 9783031314377

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

SP - 311

EP - 325

BT - Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings

A2 - Gade, Rikke

A2 - Felsberg, Michael

A2 - Kämäräinen, Joni-Kristian

PB - Springer Science and Business Media Deutschland GmbH

T2 - 23nd Scandinavian Conference on Image Analysis, SCIA 2023

Y2 - 18 April 2023 through 21 April 2023

ER -

Kristoffersson Lind S, Triebel R, Nardi L, Krueger V. Out-of-Distribution Detection for Adaptive Computer Vision. In Gade R, Felsberg M, Kämäräinen JK, editors, Image Analysis - 23rd Scandinavian Conference, SCIA 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 311-325. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-31438-4_21