Mohammadi Marzieh ; Salarpour Amir

Abstract

This paper introduces Point-GN, a novel non-parametric network for efficientand accurate 3D point cloud classification. Unlike conventional deep learningmodels that rely on a large number of trainable parameters, Point-GN leveragesnon-learnable components-specifically, Farthest Point Sampling (FPS), k-NearestNeighbors (k-NN), and Gaussian Positional Encoding (GPE)-to extract both localand global geometric features. This design eliminates the need for additionaltraining while maintaining high performance, making Point-GN particularlysuited for real-time, resource-constrained applications. We evaluate Point-GNon two benchmark datasets, ModelNet40 and ScanObjectNN, achievingclassification accuracies of 85.29% and 85.89%, respectively, whilesignificantly reducing computational complexity. Point-GN outperforms existingnon-parametric methods and matches the performance of fully trained models, allwith zero learnable parameters. Our results demonstrate that Point-GN is apromising solution for 3D point cloud classification in practical, real-timeenvironments.