High- and Low-level image component decomposition using VAEs for improved reconstruction and anomaly detection

TL;DR

This paper introduces a modified hierarchical VAE with an additional branch to better separate high- and low-level features, resulting in sharper reconstructions and improved anomaly detection in medical imaging.

Contribution

The paper proposes a novel branch in hierarchical VAEs to enhance feature separation, leading to improved image quality and anomaly detection capabilities.

Findings

Sharper and more accurate image reconstructions

Comparable or slightly improved out-of-distribution detection

Effective separation of high- and low-level features

Abstract

Variational Auto-Encoders have often been used for unsupervised pretraining, feature extraction and out-of-distribution and anomaly detection in the medical field. However, VAEs often lack the ability to produce sharp images and learn high-level features. We propose to alleviate these issues by adding a new branch to conditional hierarchical VAEs. This enforces a division between higher-level and lower-level features. Despite the additional computational overhead compared to a normal VAE it results in sharper and better reconstructions and can capture the data distribution similarly well (indicated by a similar or slightly better OoD detection performance).