Flexicubes: High-Quality Mesh Extraction
Extract triangle meshes in AI and reconstruction pipelines, yielding cleaner, better-fitting meshes for downstream tasks.
Overview
We develop advanced geometry processing tools for 3D data like meshes, point clouds, Gaussian splats, implicit surfaces, CAD geometry, and more, all fast and scalable on the GPU. Whether it comes from reconstructions, generations, or industrial workflows, real 3D data is messy and challenging—our state-of-the-art geometry pipelines unlock applications in AI, simulation, and more.
Points of Contact:
Nicholas Sharp,
Mikaela Angelina Uy,
Donglai Xiang,
Sangeetha Grama Srinivasan,
Tianchang Shen
Key Research Areas
Mesh Generation • Simulatable Geometry • Shape Analysis • 3D Representations
Featured Research Projects
PartField: 3D Feature Fields for Segmentation
A fast feedforward 3D model which segments any shape into semantic parts for understanding and manipulation.
SpaceMesh: Generating Manifold Meshes
An encoding for meshes which guarantees manifold, high-quality connectivity for generative AI architectures.
Resources & Publications
- Python Geometry Processing in NVIDIA Kaolin
- PartField: Learning 3D Feature Fields for Part Segmentation and Beyond - ICCV 2025
- SpaceMesh: A Continuous Representation for Learning Manifold Surface Meshes - SIGGRAPH 2024
- Flexible Isosurface Extraction for Gradient-Based Mesh Optimization - SIGGRAPH 2023
- Deep Marching Tetrahedra: a Hybrid Representation for High-Resolution 3D Shape Synthesis - NeurIPS 2021