Inpainting Transformer for Anomaly Detection

TL;DR

This paper introduces InTra, a self-attention based inpainting transformer that improves anomaly detection by integrating distant image information, achieving state-of-the-art results on the MVTec AD dataset.

Contribution

It proposes a novel pure self-attention inpainting transformer for anomaly detection, eliminating convolutions and effectively capturing large region context.

Findings

InTra achieves state-of-the-art detection results on MVTec AD.

InTra surpasses existing methods in segmentation performance.

The model performs well without relying on extra training data.

Abstract

Anomaly detection in computer vision is the task of identifying images which deviate from a set of normal images. A common approach is to train deep convolutional autoencoders to inpaint covered parts of an image and compare the output with the original image. By training on anomaly-free samples only, the model is assumed to not being able to reconstruct anomalous regions properly. For anomaly detection by inpainting we suggest it to be beneficial to incorporate information from potentially distant regions. In particular we pose anomaly detection as a patch-inpainting problem and propose to solve it with a purely self-attention based approach discarding convolutions. The proposed Inpainting Transformer (InTra) is trained to inpaint covered patches in a large sequence of image patches, thereby integrating information across large regions of the input image. When training from scratch, in comparison to other methods not using extra training data, InTra achieves results on par with the current state-of-the-art on the MVTec AD dataset for detection and surpassing them on segmentation.