Conditioning Latent-Space Clusters for Real-World Anomaly Classification

TL;DR

This paper introduces a method using a conditioned Variational Autoencoder to classify anomalies in urban driving scenes from high-resolution camera data, effectively separating normal and anomalous data in the latent space.

Contribution

It proposes a novel latent space conditioning approach with discrepancy maps to improve small anomaly detection in autonomous driving scenarios.

Findings

Effective separation of normal and anomalous data in latent space

Improved detection of small anomalies with discrepancy maps

High-quality image reconstruction alongside anomaly classification

Abstract

Anomalies in the domain of autonomous driving are a major hindrance to the large-scale deployment of autonomous vehicles. In this work, we focus on high-resolution camera data from urban scenes that include anomalies of various types and sizes. Based on a Variational Autoencoder, we condition its latent space to classify samples as either normal data or anomalies. In order to emphasize especially small anomalies, we perform experiments where we provide the VAE with a discrepancy map as an additional input, evaluating its impact on the detection performance. Our method separates normal data and anomalies into isolated clusters while still reconstructing high-quality images, leading to meaningful latent representations.