Patch-aware Vector Quantized Codebook Learning for Unsupervised Visual Defect Detection

TL;DR

This paper introduces a patch-aware vector quantized codebook learning method within an enhanced VQ-VAE framework, significantly improving unsupervised visual defect detection accuracy by optimizing spatial representations and code assignment.

Contribution

The paper presents a novel patch-aware dynamic code assignment scheme for VQ-VAE, enhancing defect detection by better capturing local context and spatial features.

Findings

Achieves state-of-the-art results on MVTecAD, BTAD, and MTSD datasets.

Improves defect detection accuracy by optimizing spatial code allocation.

Enhances normal-defect distinction through context-sensitive code assignment.

Abstract

Unsupervised visual defect detection is critical in industrial applications, requiring a representation space that captures normal data features while detecting deviations. Achieving a balance between expressiveness and compactness is challenging; an overly expressive space risks inefficiency and mode collapse, impairing detection accuracy. We propose a novel approach using an enhanced VQ-VAE framework optimized for unsupervised defect detection. Our model introduces a patch-aware dynamic code assignment scheme, enabling context-sensitive code allocation to optimize spatial representation. This strategy enhances normal-defect distinction and improves detection accuracy during inference. Experiments on MVTecAD, BTAD, and MTSD datasets show our method achieves state-of-the-art performance.