SoftPatch+: Fully Unsupervised Anomaly Classification and Segmentation

Abstract: Although mainstream unsupervised anomaly detection (AD) (including image-level classification and pixel-level segmentation)algorithms perform well in academic datasets, their performance is limited in practical application due to the ideal experimental setting of clean training data. Training with noisy data is an inevitable problem in real-world anomaly detection but is seldom discussed. This paper is the first to consider fully unsupervised industrial anomaly detection (i.e., unsupervised AD with noisy data). To solve this problem, we proposed memory-based unsupervised AD methods, SoftPatch and SoftPatch+, which efficiently denoise the data at the patch level. Noise discriminators are utilized to generate outlier scores for patch-level noise elimination before coreset construction. The scores are then stored in the memory bank to soften the anomaly detection boundary. Compared with existing methods, SoftPatch maintains a strong modeling ability of normal data and alleviates the overconfidence problem in coreset, and SoftPatch+ has more robust performance which is articularly useful in real-world industrial inspection scenarios with high levels of noise (from 10% to 40%). Comprehensive experiments conducted in diverse noise scenarios demonstrate that both SoftPatch and SoftPatch+ outperform the state-of-the-art AD methods on the MVTecAD, ViSA, and BTAD benchmarks. Furthermore, the performance of SoftPatch and SoftPatch+ is comparable to that of the noise-free methods in conventional unsupervised AD setting. The code of the proposed methods can be found at https://github.com/TencentYoutuResearch/AnomalyDetection-SoftPatch.

• The paper introduces SoftPatch+ as a novel framework for fully unsupervised anomaly detection and segmentation using a structured LaTeX research template.
• The template adheres to established academic formats, laying the groundwork for future detailed methodology and empirical validation.
• Despite the current absence of experimental results and comprehensive analysis, the framework offers potential to advance unsupervised anomaly classification research.

Overview of the Paper

This document is an incomplete template for a preprint submission to the journal "Nuclear Physics B." It lacks specific details such as a title, authors, and an abstract, all of which are crucial components for a complete academic paper. Although we can observe the skeletal structure of the document, including sections for a graphical abstract, highlights, keywords, and bibliography, the absence of content makes it impossible to assess the contributions or findings of this paper. Thus, without the substantive elements typically present in a research paper, such as a detailed methodology, experimental results, or theoretical analysis, this document cannot be evaluated or summarized in terms of scientific contribution or implications.

The template provided is a common structure used in academic writing, adhering to the standards of scientific documentation. The utilization of a LaTeX document class suitable for the journal indicates a well-structured presentation format, yet, further content is essential for it to serve its purpose of contributing to scholarly discourse. Such frameworks are integral for organizing information logically, aiding the process of peer review, and ensuring a consistent presentation across submissions to the journal.

Given this context, any discussion of theoretical implications, practical applications, or possible future developments stemming from this work is untenable. Moreover, without explicit numerical results or claims, the document does not offer any points of analysis or critique in its current form. An academic evaluation hinges on the data and arguments presented, which are currently nonexistent in this draft.

For colleagues or researchers looking for insights into nuclear physics advancements, this template would serve as a procedural starting point rather than a source of scientific knowledge or hypothesis testing. In essence, the document represents potential rather than an actualized contribution to the field. Therefore, awaiting the completion with substantive research content is crucial before a meaningful scholarly discussion can ensue.