RadialNet: a point cloud classification approach using local structure representation with radial basis function

Abstract

The major challenge of 3D point cloud classification using a deep neural network is to handle the naturally unordered data structure. Recently, the PointNet has achieved promising results by directly learning on point sets. However, it does not take full advantage of its local neighborhood that contains fine-grained structural information, which turns out to be helpful toward better semantic learning. To tackle this challenge, this paper proposes a new deep neural network architecture, called RadialNet, which applies radial basis function to exploit local structure representation point cloud data and then further incorporates into the conventional neural network classification framework. More specifically, the proposed architecture exploits a novel type of representation that uses evenly spaced centroids, which uses radial basis function to get weight for each input point to have order invariant and retain the global information of the point cloud. By the effective exploration of the point cloud local structure using the RadialNet, the proposed architecture achieves competitive performance on the 3D object classification benchmark ModelNet dataset.