Core Concepts
Novel approach predicts intersection points for improved surface reconstruction.
Abstract
Introduction
Surface reconstruction from point clouds is crucial in computer vision and graphics.
Traditional methods like Poisson Surface Reconstruction have limitations.
SDF-based vs. UDF-based Methods
SDF excels at smooth meshes but struggles with open surfaces.
UDF can represent open surfaces but introduces noise near the surface.
Proposed Approach
Directly predicts intersection points between line segments and implicit surfaces.
Improves reconstruction quality and eliminates artifacts in the mesh.
Methodology
Utilizes Marching Cubes for mesh generation.
Importance of accurate sign prediction and intersection positions highlighted.
Experiments
Outperforms existing methods on ShapeNet, MGN, and ScanNet datasets.
Ablation Study
Sign module and intersection module significantly impact surface quality.
Stats
|P1Q|
|P2Q|
α = |P1Q| / |P1P2|
Quotes
"Our approach demonstrates state-of-the-art performance on three datasets: ShapeNet, MGN, and ScanNet."
"Accuracy of intersection positions significantly impacts the quality of surface reconstruction."