[2H5-E-2] Machine learning: new modeling

[2H5-E-2-01] Local Feature Fitting Learning Network for Point Cloud Classification

LFFLN Convolution simulation network

〇Lu SUN¹ (1. Chiba University )

Keywords:Point cloud classification, Point cloud convolution, Nonlinear fitting

We propose a deep learning network framework to solve the classification task of three-dimensional point clouds. According to different functions, the network can be divided into the resampling block, transform block, local feature fitting block, and classification block. Unlike other classification methods based on point cloud, we try to fit local point cloud and use the fitting function as a local feature to enter the classification layer. Though simple, Local feature fitting learning network (LFFLN) is highly efficient and effective. It achieves excellent performance in the ModelNet40 without any tricks.