GramGAN: Deep 3D Texture Synthesis From 2D Exemplars

To achieve a highly realistic output conditioned on an exemplar patch, we propose a novel loss function that combines ideas from both style transfer and generative adversarial networks.