Abstract
We introduce Scan2Mesh, a novel data-driven generative approach whichtransforms an unstructured and potentially incomplete range scan into astructured 3D mesh representation. The main contribution of this work is agenerative neural network architecture whose input is a range scan of a 3Dobject and whose output is an indexed face set conditioned on the input scan.In order to generate a 3D mesh as a set of vertices and face indices, thegenerative model builds on a series of proxy losses for vertices, edges, andfaces. At each stage, we realize a one-to-one discrete mapping between thepredicted and ground truth data points with a combination of convolutional- andgraph neural network architectures. This enables our algorithm to predict acompact mesh representation similar to those created through manual artisteffort using 3D modeling software. Our generated mesh results thus producesharper, cleaner meshes with a fundamentally different structure from thosegenerated through implicit functions, a first step in bridging the gap towardsartist-created CAD models.