Symmetry-Complementary Information Fusion Self-Supervised Model for Point Cloud Completing

Self-supervised point cloud completion models rely solely on partial point clouds for training, making them more suitable for real-world point cloud completion tasks. However, this training approach limits the information available to the model, resulting in coarse completion results, which hinder the recovery of complete and fine-grained point cloud structures. To address this issue, we propose an optimization strategy that leverages symmetry information and complementary information to enhance the model's global perception capability and completion accuracy. Furthermore, to improve the model's sensitivity to local structures and enhance fine-grained completion performance, we introduce the Region-aware Chamfer Distance as a loss function. Experimental results validate the effectiveness of the proposed method.