Advancing Anomaly Detection: The IDW dataset and MC algorithm


Alexander D. J. Taylor (University of Bath), Jonathan James Morrison (Rolls-Royce Defence Aerospace), Phillip Tregidgo (University of Bristol), Neill D. F. Campbell (University of Bath)
The 35th British Machine Vision Conference
PDFPoster

Abstract

In this work we present a novel anomaly detection dataset, Industrial Defects in the Wild (IDW). IDW contains images of various industrial and household inspection processes. It contains real images with complex and varied perspectives from freely moving cameras. We show this is more challenging than the well-known MVTec dataset. We also present MultiCore (MC), a novel algorithm that obtains state-of-the-art results on the introduced IDW and popular MVtec dataset. The MC algorithm trains multiple nearest neighbour predictors, each with different hyperparameters. We propose that an ensemble is more powerful than any individual model. Synthetic anomalies are created using a novel schema intended to systematically cover as many variations as possible. The ensemble output is fed into a heatmap fusion module, which is trained in a supervised fashion using the synthetic anomalies and a perimeter-based loss function. On the popular MVTec dataset, the MC algorithm achieves P-AUC score of 0.986. On the introduced and more challenging IDW dataset, the MC algorithm achieves P-AUC of 0.889. We verify that these results are state-of-the-art by trialing the existing top fourteen anomaly detection algorithms which have code available. We release all our code.

Citation

@inproceedings{Taylor_2024_BMVC,
author    = {Alexander D. J. Taylor and Jonathan James Morrison and Phillip Tregidgo and Neill D. F. Campbell},
title     = {Advancing Anomaly Detection: The IDW dataset and MC algorithm},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0070.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection