UNSUPERVISED SEGMENTATION OF DEFORMING 3D MESHES VIA DEFORMATION-AWARE GRAPH CUTS

We propose an unsupervised method for segmenting deforming 3D mesh sequences using a deformation-aware graph cut. Our approach constructs a spatiotemporal graph and formulates segmentation as a minimum s-t cut problem. Unary costs, derived from per-vertex deformation energy, separate near-rigid parts from deforming regions, while pairwise costs enforce spatial smoothness. By iteratively applying max-flow/min-cut, the algorithm greedily extracts coherent parts without supervision. Results show the automatic partitioning of complex animations into meaningful, temporally-consistent components, validating our approach.