CAD Shape

CAD shape reconstruction from various input data (point clouds, voxel models, sketches) is a central problem in computer-aided design, aiming to create editable and manufacturable digital representations of physical objects. Current research emphasizes self-supervised learning methods, employing neural networks (including implicit field representations and those based on Constructive Solid Geometry (CSG) principles) to achieve accurate and compact model reconstructions. These advancements improve reverse engineering processes, enabling efficient digitalization of existing parts and facilitating automated manufacturing planning. Furthermore, research is actively exploring feature recognition within CAD models to streamline downstream applications like process planning and toolpath generation.