Pranoy Radhakrishnan says that if Deep Learning is Software 2.0, then Generative Adversarial Networks are Photoshop 2.0. Generative Adversarial Networks (GAN) can be used to map a semantic map of an image to a photorealistic synthesized image. Think a much nicer version of the classic “Children’s Drawings Painted Realistically”. If the introductory article piques your interest, you should read the paper, High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs. To see how it’s implemented, check out the code.