PointPCA2-RS: PointPCA2 with Resource Saving

Arthur Henrique Silva Carvalho, Pedro Garcia Freitas

DOI: 10.14209/sbrt.2025.1571150580

Evento: XLIII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2025)

Keywords: Point cloud quality assessment resource saving 3d signal processing

Abstract

This paper introduces PointPCA2-RS, a lightweight version of PointPCA2 with emphasis on low computational resources usage. PointPCA2-RS optimizes the original PointPCA2 by re-implementing it using a different programming language, structure, and specific algorithmic details. Designed to evaluate visual quality, both projects extract identical features from point cloud (PC) data. Their core functionalities are similar, including Principal Component Analysis (PCA) on local geometric data, computation of spatial and geometric descriptors, and feature pooling and aggregation. These attributes result from a programming language change, affecting architecture and performance. PointPCA2-RS adopts a modular approach opti- mized for performance and maintainability, being more suitable for large-scale or real-time applications. Experimental results demonstrate the high performance of PointPCA2-RS against PointPCA2 without sacrificing prediction accuracy. PointPCA2- RS outperforms state-of-the-art point cloud quality assessment (PCQA) metrics, offering significant improvements for PCQA field. The code of PointPCA2-RS metric is available at https: //github.com/arthurhscarvalho/pointpca2-rs.

Download