DFG-PCN: Point Cloud Completion with Degree-Flexible Point Graph

Abstract--Point cloud completion is a vital task focused on reconstructing complete point clouds and addressing the incompleteness caused by occlusion and limited sensor resolution. This limitation leads to inefficient representation and suboptimal reconstruction, especially in areas with fine-grained details or structural discontinuities. This paper proposes a point cloud completion framework called Degree-Flexible Point Graph Completion Network (DFG-PCN). It adaptively assigns node degrees using a detail-aware metric that combines feature variation and curvature, focusing on structurally important regions. We further introduce a geometry-aware graph integration module that uses Manhattan distance for edge aggregation and detail-guided fusion of local and global features to enhance representation. Extensive experiments on multiple benchmark datasets demonstrate that our method consistently outperforms state-of-the-art approaches. The point clouds obtained from real-world scenarios are frequently characterized by significant sparsity and incompleteness. These challenges arise primarily due to constraints such as restricted viewpoints, occlusion caused by object self-geometry, and the limited resolution of sensing equipment. Therefore, recovering complete point clouds is an essential downstream task, primarily aimed at preserving the observed details, inferring missing parts, and densifying sparse surfaces [8, 9]. In recent years, deep learning-based methods have been developed for point cloud completion. Notably, with the success of PointNet [10] and PointNet++ [11] in point cloud deep learning, most methods [12-15] directly generate complete point clouds based on 3D coordinates.