A progressive mesh simplification algorithm based on neural implicit representation

Abstract

Progressive mesh simplification (PM) algorithm aims to generate simplified mesh at any resolution for the input high-precision mesh, and only needs to be optimized or fitted once. Most of the existing PM algorithms are obtained based on heuristic mesh simplification algorithms, which leads to redundant storage space and poor practice-ability of the algorithm. In this article, a progressive mesh simplification algorithm based on neural implicit representation (NePM) is proposed, and NePM transforms algorithm process into an implicit continuous optimization problem through neural network and probabilistic model. NePM uses Gaussian mixture model to model high-precision mesh and samples the probabilistic model to obtain simplified meshes at different resolutions. In addition, the simplified mesh is optimized through multi-level neural network, preserving characteristics of the input high-precision mesh. Thus, the algorithm in this work lowers the memory usage of the PM and improves the practicability of the algorithm while ensuring the accuracy.