DeepCAD: A Deep Generative Network for Computer-Aided Design Models

Deep generative models of 3D shapes have received a great deal of researchinterest. Yet, almost all of them generate discrete shape representations, suchas voxels, point clouds, and polygon meshes. We present the first 3D generativemodel for a drastically different shape representation --- describing a shapeas a sequence of computer-aided design (CAD) operations. Unlike meshes andpoint clouds, CAD models encode the user creation process of 3D shapes, widelyused in numerous industrial and engineering design tasks. However, thesequential and irregular structure of CAD operations poses significantchallenges for existing 3D generative models. Drawing an analogy between CADoperations and natural language, we propose a CAD generative network based onthe Transformer. We demonstrate the performance of our model for both shapeautoencoding and random shape generation. To train our network, we create a newCAD dataset consisting of 178,238 models and their CAD construction sequences.We have made this dataset publicly available to promote future research on thistopic.