Research map / 3D vision / Face and shape
VoroMesh: Learning Watertight Surface Meshes with Voronoi Diagrams
Nissim Maruani, Roman Klokov, Maks Ovsjanikov, Pierre Alliez, Mathieu Desbrun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use VoroMesh generators instead of Marching Cubes/Dual Contouring at same grid resolution | VoroMesh optimization · Thingi32 (32^3) | Chamfer distance (x10^-5): 5.992 (DC) -> 0.791 | Improved |
| use VoroLoss over only bisectors instead of soft-Voronoi loss over all generators | VoroMesh vs VoronoiNet · Thingi32 (32^3) | Time (s): 11.6 -> 2.0 | Improved |
| learn generator positions with VoroLoss in a sparse CNN pipeline vs NDC/NMC | VoroMesh learning pipeline · ABC (32^3 input) | Watertight: 31% (NDC), 25% (NMC) -> 100% | Improved |
| train on both 32^3 and 64^3 grids (VoroMesh32+64) instead of 32^3 only (ablation) | VoroMesh · ABC (32^3 input) | Chamfer distance (x10^-5): 13.15 -> 2.23 | Improved |
| train on both grid resolutions instead of one, evaluated on unseen Thingi32 at 64^3 input (ablation) | VoroMesh · Thingi32 generalization (64^3) | Chamfer distance (x10^-5): 0.868 -> 1.031 | Hurt |
| VoroMesh vs NMC at higher resolution on Thingi32 generalization | VoroMesh · Thingi32 generalization (128^3) | Chamfer distance (x10^-5): 0.642 (NMC) vs 0.659 | Hurt |
| VoroMesh vs NMC on normal consistency in ABC learning task | VoroMesh · ABC (64^3 input) | Normal consistency: 0.980 (NMC) vs 0.966 | Hurt |
| replace tanh offset bound with mild max-offset regularization (ablation) | VoroMesh network · ABC training | training behavior | Improved |
| compare to DMT with direct optimization and GT SDF initialization | VoroMesh · Thingi32 (64^3) | Chamfer distance: 0.712 -> 0.645 | Improved |
Later papers that compared with this method
2 later accepted papers report VoroMesh as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DMesh++: An Efficient Differentiable Mesh for Complex Shapes (ICCV 2025 · better on 2 of 3 shared results)
- PoNQ: a Neural QEM-based Mesh Representation (CVPR 2024 · better on 1 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PoNQ: a Neural QEM-based Mesh Representation (CVPR 2024 · similarity 0.87)
- Voint Cloud: Multi-View Point Cloud Representation for 3D Understanding (ICLR 2023 · similarity 0.87)
- Voronoi Boundary Classification: A High-Dimensional Geometric Approach via Weighted Monte Carlo Integration (ICML 2019 · similarity 0.86)
- ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction (NeurIPS 2025 · similarity 0.85)
- Ghost on the Shell: An Expressive Representation of General 3D Shapes (ICLR 2024 · similarity 0.85)
- PolyhedronNet: Representation Learning for Polyhedra with Surface-attributed Graph (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
