Anomaly detection (AD) is a vital process in industrial applications, used to discover unexpected events within the input data. This process is commonly applied to investigate images and detect defects, however it is especially difficult attributable to the complexity of the defects, which will be extremely tiny and hard to gather. Unsupervised AD is a key tool in handling this complexity.

Most previous unsupervised AD methods depend on clean training data to extract nominal features and compare them with anomalous features. Therefore, having noisy data (which is inevitable in real-world settings) can significantly affect the performance of those models. In this research paper, the authors have focused on the importance of studying noisy data problems in unsupervised AD and have introduced a novel algorithm named which utilizes the outlier factor to realize higher noise robustness.

Previous AD methods, similar to PatchCore and CFA, consist of three fundamental processes – feature extraction, coreset selection with a memory bank (a big set of vectors describing what normal image patches appear to be), and anomaly detection. Researchers from the Southern University of Science and Technology and Tencent introduced SoftPatch as an architecture much like these methods. However, it first filters the noisy data using a noise discriminator before the coreset construction process, thereby softening the search process.

SoftPatch distinguishes the noise in the info on the patch level at each position of the feature map. With a rise in training images, the feature memory becomes infeasible for differentiating noise, and thus, SoftPatch groups all features by position and counts their outlier rating. Subsequently, the scores are aggregated to find out the noise patches, after which the features with probably the most noise are removed. After this, the anomaly scores are calculated and grouped by noise level. This process considers the local relationship around the closest node, which increases its robustness.

The researchers evaluated their work in various noise scenes, and the outcomes display that SoftPatch outperforms state-of-the-art AD methods on the MVTec Anomaly Detection benchmark. Moreover, SoftPatch also achieved optimal results on the BTAD dataset, highlighting its effectiveness.

In conclusion, this paper emphasizes the importance of investigating noisy data in unsupervised AD. It is certainly one of the primary works to concentrate on this practical problem, which is commonly missed. SoftPatch’s impressive performance in various noise scenes provides a brand new view for further research and has the potential to further improve the efficiency and performance of commercial inspection systems.

Check out the Paper and GithubAll credit for this research goes to the researchers of this project. Also, don’t forget to follow us on Twitter. Join our Telegram Channel, Discord Channel, and LinkedIn Group.

Don’t Forget to hitch our 39k+ ML SubReddit

This article was originally published at