Survey on Unsupervised Domain Adaptation for Semantic Segmentation for Visual Perception in Automated Driving

Deep neural networks (DNNs) have proven their capabilities in the past years and play a significant role environment perception for challenging application of automated driving. They are employed tasks such as detection, semantic segmentation, sensor fusion. Despite tremendous research efforts, several issues still need to be addressed that limit applicability DNNs The bad generalization unseen domains is major problem on way safe, large-scale application, because manual annotation new costly, particularly segmentation. For this reason, methods required adapt without labeling effort. This task termed unsupervised domain adaptation (UDA). While different shifts challenge DNNs, shift between synthetic real data particular importance driving, it allows use simulation environments DNN training. We present an overview current state art field. categorize explain approaches UDA. number considered publications larger than any other survey topic. also go far beyond description UDA state-of-the-art, we quantitative comparison point out latest trends conduct critical analysis state-of-the-art highlight promising future directions. With survey, aim facilitate further encourage scientists exploit novel