Detecting Adversarial Attacks in Semantic Segmentation via Uncertainty Estimation: A Deep Analysis | Synapse