3D Synthetic Image Training and Testing Data for Autonomous Driving

In this paper, we propose generating training and testing images for autonomous driving using the CARLA simulator. Several experiments have been performed to prove the effectiveness of training on synthetic data, including those on mixed datasets including both data generated from CARLA and real-world data. The result of such experiments showed that synthetic data generated using the CARLA simulator was effective as training and testing data for on-road object detection.