PCTN: Point cloud data transformation network

In point cloud classification tasks, efficiently extracting point cloud data feature has always been a challenging problem. Based on the characteristics of the point cloud data distribution, the point cloud from different parts contains distinct feature information. Therefore, they cannot be treated equally. In this work, we propose a novel method called PCTN, which primarily consists of two modules: the Planar-Contour (PC) module and the Planar-Contour Attention (PCA) module. The former can partition the point cloud data into different parts based on its spatial structural distribution. Subsequently, the latter performs feature extraction and fusion on the corresponding parts by applying attention, resulting in feature that not only reduce redundancy but also possess distinctiveness. We conduct experiments on different datasets, achieving an experimental accuracy of 93.4% on the ModelNet40 dataset and 87.7%±0.4 on the ScanObjectNN dataset, demonstrating the effectiveness of our model.