UCSG-Net -- Unsupervised Discovering of Constructive Solid Geometry Tree
This addresses the limitation of existing supervised CSG methods in 3D shape reconstruction for applications like CAD design, though it is incremental as it builds on prior CSG frameworks.
The paper tackles the problem of reconstructing non-convex 3D shapes by proposing UCSG-Net, an unsupervised method that discovers constructive solid geometry (CSG) parse trees without requiring supervision, achieving interpretable results usable in CAD software.
Signed distance field (SDF) is a prominent implicit representation of 3D meshes. Methods that are based on such representation achieved state-of-the-art 3D shape reconstruction quality. However, these methods struggle to reconstruct non-convex shapes. One remedy is to incorporate a constructive solid geometry framework (CSG) that represents a shape as a decomposition into primitives. It allows to embody a 3D shape of high complexity and non-convexity with a simple tree representation of Boolean operations. Nevertheless, existing approaches are supervised and require the entire CSG parse tree that is given upfront during the training process. On the contrary, we propose a model that extracts a CSG parse tree without any supervision - UCSG-Net. Our model predicts parameters of primitives and binarizes their SDF representation through differentiable indicator function. It is achieved jointly with discovering the structure of a Boolean operators tree. The model selects dynamically which operator combination over primitives leads to the reconstruction of high fidelity. We evaluate our method on 2D and 3D autoencoding tasks. We show that the predicted parse tree representation is interpretable and can be used in CAD software.