Hsu-Shen Liu*, Huang-Ru Liao*, Jie-En Yao*, Li-Yuan Tsao*, Shan-Ya Yang*, Ting-Hsuan Liao*, Tzu-Wen Wang* (in alphabetical order)
This work focuses on applying Unsupervised Domain Adaptation for Semantic Segmentation Task (UDA-SST) on Autonomous Driving. We proposed some different ways to address this problem, including Consistency Learning, Edge Prediction, Color Quantization, Gray-World Algorithm, and each of them can achieve state-of-the-art performance. We further combine our UDA-SST methods and Reinforcement Learning (RL) to train a self-driving RL agent in the simulation environment and conduct real-world experiments. We show that our RL agent can perform well in both simulation and real-world environments with the proposed methods.