This project sets out to investigate if semantic segmentation can be used to improve anomaly detection. The problem analysis is centered around finding the best practices for using a dataset on each system without ground truth for both systems available. To achieve this, an analysis is made on segmentation results from models on the anomaly detection dataset UCHK Avenue, and the semantic segmentation dataset Cityscapes is manually annotated with frame specific anomalies. The system used for the anomaly detection is a General adversarial network predicting future frames from spatial and temporal constraints. Testing on the datasets showed that segmented frames can achieve accuracy with the system competitive with RGB frames given a high enough segmentation accuracy.