NeuroMorph: Unsupervised Shape Interpolation and Correspondence in One Go

Figure 1: Our NeuroMorph neural network takes as input two meshes (left and right) and produces in one go (i.e. in a single feed-forward pass) a continuous interpolation and point-to-point correspondence between them (color coded). The interpolation, expressed as a displacement field, changes the pose of the source shape while preserving its identity.