Scan2Mesh: From Unstructured Range Scans to 3D Meshes
This addresses the challenge of automating 3D mesh creation from scans for applications like CAD modeling, though it is incremental in bridging the gap towards artist-created models.
The paper tackles the problem of converting unstructured and incomplete range scans into structured 3D meshes, introducing Scan2Mesh, a generative neural network that outputs indexed face sets with sharper, cleaner results compared to implicit function methods.
We introduce Scan2Mesh, a novel data-driven generative approach which transforms an unstructured and potentially incomplete range scan into a structured 3D mesh representation. The main contribution of this work is a generative neural network architecture whose input is a range scan of a 3D object 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, the generative model builds on a series of proxy losses for vertices, edges, and faces. At each stage, we realize a one-to-one discrete mapping between the predicted and ground truth data points with a combination of convolutional- and graph neural network architectures. This enables our algorithm to predict a compact mesh representation similar to those created through manual artist effort using 3D modeling software. Our generated mesh results thus produce sharper, cleaner meshes with a fundamentally different structure from those generated through implicit functions, a first step in bridging the gap towards artist-created CAD models.