DFR: Deep Feature Reconstruction for Unsupervised Anomaly Segmentation

TL;DR

This paper introduces a novel unsupervised anomaly segmentation method that uses multi-scale regional features from pre-trained networks and a deep autoencoder to detect small, confined anomalies in images effectively.

Contribution

It proposes a multi-scale regional feature generator combined with a deep autoencoder for improved unsupervised anomaly segmentation, especially for small regions.

Findings

Outperforms existing methods on benchmark datasets

Effective at detecting small and confined anomalies

Fast and computationally efficient

Abstract

Automatic detecting anomalous regions in images of objects or textures without priors of the anomalies is challenging, especially when the anomalies appear in very small areas of the images, making difficult-to-detect visual variations, such as defects on manufacturing products. This paper proposes an effective unsupervised anomaly segmentation approach that can detect and segment out the anomalies in small and confined regions of images. Concretely, we develop a multi-scale regional feature generator that can generate multiple spatial context-aware representations from pre-trained deep convolutional networks for every subregion of an image. The regional representations not only describe the local characteristics of corresponding regions but also encode their multiple spatial context information, making them discriminative and very beneficial for anomaly detection. Leveraging these descriptive regional features, we then design a deep yet efficient convolutional autoencoder and detect anomalous regions within images via fast feature reconstruction. Our method is simple yet effective and efficient. It advances the state-of-the-art performances on several benchmark datasets and shows great potential for real applications.