Bijective upsampling and learned embedding for point clouds correspondences

Giulio Viganò, Simone Melzi

Abstract: Highlights•We propose a new method for point cloud data-driven correspondences.•We theoretically prove some properties of a new conversion method for the Functional Map method.•We design a novel pipeline which merges the learning of embeddings for point cloud matching a refinement algorithm for bijective correspondences.•We train PointNet++ to compute embedding for point-cloud matching.