SoftPatch: A Memory-Based Unsupervised Anomaly Detection AD Method that Efficiently Denoises the Data at the Patch Level

Anomaly detection (AD) is a crucial process in industrial applications, used to identify unexpected events in the input data. This process is often applied to analyze images and detect defects, but it is particularly challenging due to the complexity of the defects, which can be extremely tiny and hard to collect. Unsupervised AD is a key tool in handling this complexity.

Most previous unsupervised AD methods rely 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 these models. In this research paper, the authors have focused on the significance of studying noisy data problems in unsupervised AD and have introduced a novel algorithm named SoftPatch, which utilizes the outlier factor to achieve better noise robustness.

Previous AD methods, such as PatchCore and CFA, consist of three main processes – feature extraction, coreset selection with a memory bank (a large set of vectors describing what normal image patches look like), and anomaly detection. Researchers from the Southern University of Science and Technology and Tencent introduced SoftPatch as an architecture similar to 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 data at the patch level at each position of the feature map. With an increase in training images, the feature memory becomes infeasible for differentiating noise, and thus, SoftPatch groups all features by position and counts their outlier score. Subsequently, the scores are aggregated to determine the noise patches, after which the features with 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 nearest node, which increases its robustness.

The researchers evaluated their work in various noise scenes, and the results demonstrate 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 one of the first works to focus on this practical problem, which is often overlooked. SoftPatch’s impressive performance in various noise scenes provides a new view for further research and has the potential to further improve the efficiency and performance of industrial 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.

If you like our work, you will love our newsletter..

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

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.

🐝 Join the Fastest Growing AI Research Newsletter Read by Researchers from Google + NVIDIA + Meta + Stanford + MIT + Microsoft and many others...