Texture Generation Using a Graph Generative Adversarial Network and Differentiable Rendering

Novel photo-realistic texture synthesis is an important task for generating novel scenes, including asset generation 3D simulations. However, to date, these methods predominantly generate textured objects in 2D space. If we rely on object generation, then need make a computationally expensive forward pass each time change the camera viewpoint or lighting. Recent work that can textures requires component segmentation acquire. In this work, present conditional generative architecture call graph adversarial network (GGAN) by learning information unsupervised way. framework, do not whenever lighting changes, and part training, yet model generalize unseen meshes appropriate textures. We compare approach against state-of-the-art demonstrate GGAN obtains significantly better quality (according Fréchet inception distance). release our source code as open ( https://github.com/ml4ai/ggan ).