Partial transport for point-cloud registration (Q6980462)

This paper addresses the challenging problem of non-rigid point-cloud registration under noise, occlusion, and partial overlap by proposing a unified framework based on optimal partial transport (OPT). The overarching goal is to overcome the mass-preservation limitation of classical optimal transport and to provide a mathematically rigorous yet computationally efficient methodology for partial and robust registration. By combining parametric deformation models (RBF and TPS) with optimal partial transport -- and introducing a sliced OPT variant to mitigate computational complexity -- the authors tackle both theoretical and algorithmic challenges inherent in large-scale non-rigid alignment. The approach is particularly relevant in applications such as robotics, medical imaging, and computer vision, where partial visibility and non-rigid deformations are intrinsic. The presentation is clear and well-structured, with a comprehensive review of related methods and a careful exposition of both optimal transport background and algorithmic implementation. Overall, the paper contributes a conceptually coherent and practically meaningful advancement to the interplay between optimal transport theory and geometric data analysis.