Fully Convolutional Cross-Scale-Flows for Image-based Defect Detection

TL;DR

This paper introduces a novel fully convolutional cross-scale normalizing flow model for image-based defect detection that effectively handles fine-grained, global, and local image features, enabling accurate image-level and defect localization without prior defect samples.

Contribution

The paper presents a new CS-Flow model that jointly processes multi-scale features with normalizing flows, improving defect detection and localization in industrial images without needing defective training samples.

Findings

Achieved 100% AUROC on 4 out of 15 classes in benchmarks.

Sets a new state-of-the-art in image-level defect detection.

Enables defect localization through interpretable latent space.

Abstract

In industrial manufacturing processes, errors frequently occur at unpredictable times and in unknown manifestations. We tackle the problem of automatic defect detection without requiring any image samples of defective parts. Recent works model the distribution of defect-free image data, using either strong statistical priors or overly simplified data representations. In contrast, our approach handles fine-grained representations incorporating the global and local image context while flexibly estimating the density. To this end, we propose a novel fully convolutional cross-scale normalizing flow (CS-Flow) that jointly processes multiple feature maps of different scales. Using normalizing flows to assign meaningful likelihoods to input samples allows for efficient defect detection on image-level. Moreover, due to the preserved spatial arrangement the latent space of the normalizing flow is interpretable which enables to localize defective regions in the image. Our work sets a new state-of-the-art in image-level defect detection on the benchmark datasets Magnetic Tile Defects and MVTec AD showing a 100% AUROC on 4 out of 15 classes.