Neural Stochastic Screened Poisson Reconstruction

TL;DR

This paper introduces a neural network-based method for surface reconstruction from point clouds that quantifies uncertainty and integrates seamlessly into 3D scanning workflows, improving reconstruction quality and decision-making.

Contribution

It presents a novel neural approach to surface reconstruction that explicitly models uncertainty and enhances the 3D scanning pipeline integration.

Findings

Effective uncertainty quantification in surface reconstruction.

Improved reconstruction accuracy over existing methods.

Seamless integration into 3D scanning workflows.

Abstract

Reconstructing a surface from a point cloud is an underdetermined problem. We use a neural network to study and quantify this reconstruction uncertainty under a Poisson smoothness prior. Our algorithm addresses the main limitations of existing work and can be fully integrated into the 3D scanning pipeline, from obtaining an initial reconstruction to deciding on the next best sensor position and updating the reconstruction upon capturing more data.