CHATGPT FOR POINT CLOUD 3D OBJECT PROCESSING

Abstract

Large-scale pretrained language models have been a revolution in human-machine communication. Recently, such language models also generate code for required tasks. The objective of this work is to evaluate the functionality of the codes generated by ChatGPT (version 15-Dec-2022) for point cloud processing. The programming language selected for the test was MATLAB due to the extensive use in prototyping and toolboxes for Computer Vision and LiDAR. Using the Question-Answer system, the ChatGPT was asked for codes to calculate surface normals, curvature, eigenvalues, and eigenfeatures, with specific parameters and outputs. The provided codes were compiled and executed. The results show that ChatGPT can generate functional code for very specific and short applications, however, it is not capable of generating large code involving the correct use of loops, indexes, or equations.