Structure-Aware Graph Construction for Point Cloud Segmentation with Graph Convolutional Networks

K-nearest neighbors (KNN) algorithm has been widely adopted to construct graph convolutional networks (GCNs) for point cloud segmentation. However, the L2 norm cannot discriminate multi-dimensional structures within point clouds. In this paper, we propose a novel structure-aware graph construction for point clouds that compensates the L2 norm with per-dimension differences of the signal. The proposed method dynamically calculates the similarity ratio to determine the dimension-based proximity of pair of points. Consequently, it improves spatial and spectral GCNs with the capability to aggregate information from relevant neighbors for point cloud segmentation. As a model-agnostic method, it can be seamlessly embedded into arbitrary GCN model during graph construction phase. Experimental results demonstrate that the proposed method can improve classification accuracy around the jointed areas of objects.