Discrepancy Scaling for Fast Unsupervised Anomaly Localization

Juha Mylläri; Jukka K Nurminen

doi:10.1109/COMPSAC57700.2023.00042

Discrepancy Scaling for Fast Unsupervised Anomaly Localization

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

Abstract

Computer vision systems can automatically find and segment anomalies in images even without ever seeing anomalous observations during training. Many methods for such unsupervised anomaly detection (AD) and localization (AL) tasks have been introduced in recent years, but the most accurate methods tend to be computationally heavy. In this paper, we propose Discrepancy Scaling, a method that significantly improves the accuracy of a very fast AD and AL approach called Student-Teacher Feature Pyramid Matching. We show that with Discrepancy Scaling, even a small, mobile-friendly convolutional neural network can perform well on AD and AL tasks.

Original language	English
Title of host publication	2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)
Number of pages	6
Publisher	IEEE
Publication date	Aug 2023
Pages	274-279
ISBN (Electronic)	979-8-3503-2697-0
DOIs	https://doi.org/10.1109/COMPSAC57700.2023.00042
Publication status	Published - Aug 2023
MoE publication type	A4 Article in conference proceedings
Event	Annual Computers, Software, and Applications Conference - Torino, Italy Duration: 26 Jun 2023 → 30 Jun 2023 Conference number: 47

Fields of Science

113 Computer and information sciences
Anomaly detection
Anomaly localization
Deep learning
Unsupervised learning

Access to Document

10.1109/COMPSAC57700.2023.00042

Discrepancy_Scaling_acceptedAccepted author manuscript, 1.43 MBLicence: Unspecified

1 Finished
1 Active

IML4E: Industrial Machine Learning for Enterprises
Nurminen, J. K. (Project manager), Hu, R. Y. (Participant), Hussain, Z. (Participant), Kemell, K.-K. (Participant), Kivimäki, J. I. (Participant), Laisi, J. S. (Participant), Lehtonen, A. I. (Participant), Louhi, J. J. (Participant), Luo, Y. (Participant), Lwakatare, L. E. M. (Participant), Muiruri, D. (Participant), Myllyaho, L. (Participant), Mylläri, J. (Participant), Mylläri, J. (Participant), Raatikainen, M. (Participant), Rantanen, R. J. (Participant), Rensing, F. (Participant), Salmenperä, I. E. (Participant), Samarasekara, H. (Participant), Siilasjoki, N. J. (Participant), Stirbu, V. A. (Participant), Vidanapathirana, J. C. (Participant), Myllyaho, L. (Participant) & Raatikainen, M. (Participant)
Innovaatiorahoituskeskus Business Finland
01/05/2021 → 28/02/2025
Project: Business Finland
IVVES: Industrial-grade Verification and Validation of Evolving Systems
Mikkonen, T. (Project manager), Nurminen, J. K. (Project manager), Becker, L. V. A. (Participant), Halvari, T. P. (Participant), Hussain, Z. (Participant), Kauhanen, E. O. (Participant), Kramar, V. T. (Participant), Laanti, T. M. (Participant), Mikkonen, T. (Participant), Muiruri, D. (Participant), Myllyaho, L. (Participant), Mylläri, J. (Participant), Raatikainen, M. (Participant), Salmenperä, I. E. (Participant), Steenari, J. P. (Participant), Stirbu, V. A. (Participant), Valjakka, J. H. (Participant), Hussain, Z. (Participant), Myllyaho, L. (Participant) & Raatikainen, M. (Participant)
Business Finland Oy
01/01/2020 → 31/03/2024
Project: Business Finland

Cite this

@inproceedings{166530a21d4e4e02a96c6a2b8276e752,

title = "Discrepancy Scaling for Fast Unsupervised Anomaly Localization",

abstract = "Computer vision systems can automatically find and segment anomalies in images even without ever seeing anomalous observations during training. Many methods for such unsupervised anomaly detection (AD) and localization (AL) tasks have been introduced in recent years, but the most accurate methods tend to be computationally heavy. In this paper, we propose Discrepancy Scaling, a method that significantly improves the accuracy of a very fast AD and AL approach called Student-Teacher Feature Pyramid Matching. We show that with Discrepancy Scaling, even a small, mobile-friendly convolutional neural network can perform well on AD and AL tasks.",

keywords = "113 Computer and information sciences, Anomaly detection, Anomaly localization, Deep learning, Unsupervised learning",

author = "Juha Myll{\"a}ri and Nurminen, {Jukka K}",

year = "2023",

month = aug,

doi = "10.1109/COMPSAC57700.2023.00042",

language = "English",

pages = "274--279",

booktitle = "2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)",

publisher = "IEEE",

address = "United States",

note = "Annual Computers, Software, and Applications Conference, COMPSAC ; Conference date: 26-06-2023 Through 30-06-2023",

}

TY - GEN

T1 - Discrepancy Scaling for Fast Unsupervised Anomaly Localization

AU - Mylläri, Juha

AU - Nurminen, Jukka K

N1 - Conference code: 47

PY - 2023/8

Y1 - 2023/8

N2 - Computer vision systems can automatically find and segment anomalies in images even without ever seeing anomalous observations during training. Many methods for such unsupervised anomaly detection (AD) and localization (AL) tasks have been introduced in recent years, but the most accurate methods tend to be computationally heavy. In this paper, we propose Discrepancy Scaling, a method that significantly improves the accuracy of a very fast AD and AL approach called Student-Teacher Feature Pyramid Matching. We show that with Discrepancy Scaling, even a small, mobile-friendly convolutional neural network can perform well on AD and AL tasks.

AB - Computer vision systems can automatically find and segment anomalies in images even without ever seeing anomalous observations during training. Many methods for such unsupervised anomaly detection (AD) and localization (AL) tasks have been introduced in recent years, but the most accurate methods tend to be computationally heavy. In this paper, we propose Discrepancy Scaling, a method that significantly improves the accuracy of a very fast AD and AL approach called Student-Teacher Feature Pyramid Matching. We show that with Discrepancy Scaling, even a small, mobile-friendly convolutional neural network can perform well on AD and AL tasks.

KW - 113 Computer and information sciences

KW - Anomaly detection

KW - Anomaly localization

KW - Deep learning

KW - Unsupervised learning

U2 - 10.1109/COMPSAC57700.2023.00042

DO - 10.1109/COMPSAC57700.2023.00042

M3 - Conference contribution

SP - 274

EP - 279

BT - 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)

PB - IEEE

T2 - Annual Computers, Software, and Applications Conference

Y2 - 26 June 2023 through 30 June 2023

ER -

Discrepancy Scaling for Fast Unsupervised Anomaly Localization

Abstract

Fields of Science

Access to Document

Projects

IML4E: Industrial Machine Learning for Enterprises

IVVES: Industrial-grade Verification and Validation of Evolving Systems

Cite this