Self-Supervised Training with Autoencoders for Visual Anomaly Detection