VFS-Net: A Vertex and Face-Based Deep Learning Pipeline for Three-Dimensional Model Segmentation and Analysis
Year: 2026
Type: preprint
Abstract: This study presents a pipeline for the segmentation and analysis of dynamic, three-dimensional models using VFS-Net, a dual vertex- and face-based pipeline. This framework demonstrates favorable segmentation and classification metrics for both biological and synthetic 3D model types, demonstrating broad applicability. Benchmarking across diverse datasets demonstrates that VFS-Net remains robust against variations in model complexity and mesh density. Traditional machine learning methods require structured grids or volumetric data and are not optimized to handle data points that vary broadly based on scale and surface area. The VFS-Net pipeline was designed to operate on 3D models obtained using structured-light scanning, enabling the accurate capture of complex surface geometrics and topological features. Future extensions of this framework will further integrate two-dimensional data, extending its biomedical research relevance.
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7133746387