Figure 1: Overview: Existing 3D representations discretize
the output space differently: (a) spatially in voxel represen-
tations, (b) in terms of predicted points, and (c) in terms of
vertices for mesh representations. In contrast, (d) we pro-
pose to consider the continuous decision boundary of a clas-
sifier f (e.g., a deep neural network) as a 3D surface which
θ
allows to extract 3D meshes at any resolution.