Enhancing Performance of 3D Point Completion Network using Consistency Loss

Kevin Tirta Wijaya, Christofel Rio Goenawan, Seung-Hyun Kong

Abstract: Highlights•Introduction of Completion Consistency loss: A novel loss function for Point Cloud Completion Networks (PCCNs) to address the one-to-many problem.•Compatibility with Existing Networks: The proposed consistency loss can be seamlessly integrated into existing PCCNs without any modification of their design.•Improved Network Performance and Efficiency: The proposed consistency loss significantly improved the performance of PCCNs, enabling simper network to achieve comparable to more complex networks.•Point Completion Network trained using proposed consistency loss on challenging MVP dataset achieve state-of-the-art performance for MVP dataset•Enhanced Generalization Capability: The proposed consistency loss also enhanced the network’s ability to generalize to previously unseen objects.