Improving Semantic Segmentation For Autonomous Vehicles Using Synthetic Images