EditVAE: Unsupervised Parts-Aware Controllable 3D Point Cloud Shape Generation

This paper tackles the problem of parts-aware point cloud generation. Unlike existing works which require to be segmented into parts a priori, our editing and generation are performed in an unsupervised manner. We achieve this with simple modification Variational Auto-Encoder yields joint model itself along schematic representation it as combination shape primitives. In particular, we introduce latent can decomposed disentangled for each part shape. These turn both primitive representation, standardising transformation canonical coordinate system. The dependencies between transformations preserve spatial manner that allows meaningful editing. addition flexibility afforded by inductive bias introduced modeling approach state-of-the-art experimental results on ShapeNet dataset.