A deep learning model developed by NVIDIA Research turns rough doodles into highly realistic scenes using generative adversarial networks (GANs).