A deep learning model developed by NVIDIA Research uses GANs to turn segmentation maps into lifelike images with breathtaking ease.