The quantum accelerated PointNet algorithm

Abstract

Point cloud modeling is one of the most common types of 3D modeling, and the PointNet algorithm is an effective point cloud classification and segmentation algorithm. We propose the quantum accelerated PointNet algorithm. The proposed algorithm uses quantum computing to realize the three convolutional layers of the PointNet algorithm and uses classical computers to realize the pooling layers, the fully connected layers, and other parts. For a point cloud with [Formula: see text] points and coordinate values up to [Formula: see text], when performing the computation of the convolution layer of the PointNet algorithm with convolution kernel weights up to [Formula: see text], our algorithm changes the computational complexity from [Formula: see text] on an electronic computer to [Formula: see text] after the quantum computing acceleration. The quantum accelerated PointNet algorithm proposed in this study changes the variables of the polynomial of computational complexity from [Formula: see text] to the product [Formula: see text], and completely removes the effect of the parameter [Formula: see text], which is positively related to the number of points. Therefore, We can conclude that the quantum accelerated PointNet algorithm achieves a certain speedup compared to the classical PointNet algorithm.