Variational Autoencoder for Anomaly Detection: A Comparative Study

TL;DR

This study compares different Variational Autoencoder architectures for anomaly detection, highlighting ViT-VAE's superior performance and emphasizing the importance of diverse datasets for robust benchmarking.

Contribution

It provides a comprehensive comparison of VAE variants in anomaly detection and introduces the use of the MiAD dataset for more robust evaluation.

Findings

ViT-VAE outperforms other architectures in various scenarios.

VAE-GRF requires careful hyperparameter tuning.

Using MiAD dataset improves benchmarking robustness.

Abstract

This paper aims to conduct a comparative analysis of contemporary Variational Autoencoder (VAE) architectures employed in anomaly detection, elucidating their performance and behavioral characteristics within this specific task. The architectural configurations under consideration encompass the original VAE baseline, the VAE with a Gaussian Random Field prior (VAE-GRF), and the VAE incorporating a vision transformer (ViT-VAE). The findings reveal that ViT-VAE exhibits exemplary performance across various scenarios, whereas VAE-GRF may necessitate more intricate hyperparameter tuning to attain its optimal performance state. Additionally, to mitigate the propensity for over-reliance on results derived from the widely used MVTec dataset, this paper leverages the recently-public MiAD dataset for benchmarking. This deliberate inclusion seeks to enhance result competitiveness by alleviating the impact of domain-specific models tailored exclusively for MVTec, thereby contributing to a more robust evaluation framework. Codes is available at https://github.com/endtheme123/VAE-compare.git.