High- and Low-level image component decomposition using VAEs for improved reconstruction and anomaly detection
This is an incremental improvement for medical image analysis, addressing specific limitations in VAE-based methods.
The paper tackled the problem of VAEs producing blurry images and lacking high-level features by introducing a new branch to conditional hierarchical VAEs to separate high- and low-level features, resulting in sharper reconstructions and similar or slightly better out-of-distribution detection performance.
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).