http://carla.org/

I wonder how easy it is to translate the scene segmentation and perception from simulated Carla data to the real world, and vice versa? I think in lecture Kayvon said it was an area of ongoing research - would love to learn more.