Abstract
Signed distance field (SDF) is a prominent implicit representation of 3Dmeshes. Methods that are based on such representation achieved state-of-the-art3D shape reconstruction quality. However, these methods struggle to reconstructnon-convex shapes. One remedy is to incorporate a constructive solid geometryframework (CSG) that represents a shape as a decomposition into primitives. Itallows to embody a 3D shape of high complexity and non-convexity with a simpletree representation of Boolean operations. Nevertheless, existing approachesare supervised and require the entire CSG parse tree that is given upfrontduring the training process. On the contrary, we propose a model that extractsa CSG parse tree without any supervision - UCSG-Net. Our model predictsparameters of primitives and binarizes their SDF representation throughdifferentiable indicator function. It is achieved jointly with discovering thestructure of a Boolean operators tree. The model selects dynamically whichoperator combination over primitives leads to the reconstruction of highfidelity. We evaluate our method on 2D and 3D autoencoding tasks. We show thatthe predicted parse tree representation is interpretable and can be used in CADsoftware.