3D shape models are naturally parameterized using vertices and faces, i.e., composed of polygons forming a surface. However, current 3D learning paradigms for predictive and generative tasks using convolutional neural networks focus on a...