This is a Python implementation of the paper "Robust Reconstruction of Watertight 3D Models from Non-uniformly Sampled Point Clouds Without Normal Information" by A. Hornung and L. Kobbelt.
The goal is to create a watertight mesh from a point cloud.
The core idea is that the watertight mesh is the the minimum of a 3D distance function ϕ to the points.
The paper proposes a minimum solver that solves the minimum of that distance function ϕ by using a max-flow min-cut through a voxel grid.
There are online interactive renders at: https://mickare.github.io/Robust-Reconstruction-of-Watertight-3D-Models/
- A low resolution voxel grid is filled via a point cloud.
- A boolean crust is created by dilation & flood-filling until a watertight voxel crust is created.
- The float distance values ϕ of each crust voxel to the model point cloud is computed via diffusion of the point cloud.
- A weighted graph of ϕ is created from the outer crust surface to the inner crust surface. The Maxflow-Mincut of this graph is the minimum of ϕ.
- Either:
- Repeat from 3. in a higher resolution by dilating the mincut to a new crust
- Finally extract and smooth a mesh from the mincut
Python 3.8
Conda environment that has numba installed.
- Installation numba
- or
conda install --file requirements.txt
Some files are not in the official conda channels, so they need to be installed via pip.
pip install -r requirements.txt
Run:
main.pyfor running the reconstruction and show each step in the browserexport.py. for exporting a web page with each reconstruction step
Change the following line in either main.py or export.py.
example = Example.BunnyFixed
| Name | Description | Source | License |
|---|---|---|---|
| Bunny | Stanford Bunny | https://graphics.stanford.edu/data/3Dscanrep/ | Free but acknowledge required. |
| BunnyFixed | The Bunny with addition points | https://graphics.stanford.edu/data/3Dscanrep/ | Free but acknowledge required. |
| Dragon | Stanford Dragon | https://graphics.stanford.edu/data/3Dscanrep/ | Free but acknowledge required. |
| Camel | From another paper | https://people.csail.mit.edu/sumner/research/deftransfer/data.html | De Espona model library |
| Cat | Free cat model | https://free3d.com/3d-model/lowpoly-cat-rigged-run-animation-756268.html | Personal Use License |
| Dog | Free dog model | https://free3d.com/3d-model/low-poly-german-shepherd-dog-26963.html | Personal Use License |
You can use your own models. This project has model loader implemented for .ply and .pts files.
Point clouds (like Cat and Dog) can be generated from existing wavefront .obj meshes.
We split the voxel space into chunks, which we stored in a flat dict which we call ChunkGrid.
Both Chunk and ChunkGrid have a fill value which is used in the case of a filled or missing chunk.
The standard math (+, -, /, *, ...) and logical (&, ^, |, ...) operators are implemented for both Chunk
and ChunkGrid, and are forwarded to the internal numpy arrays or fill values.
The chunk size N³ (e.g. 16x16x16) can be configured in main.py or export.py.
You can find the data structure in data/chunks.py.
Plotly is used for plotting the 3D point clouds, meshes and voxels.
The voxel mesh is created in render/voxel_render.py.
It reduces the number of triangle faces by merging neighboring voxel faces.
The code in this repository is licensed under the MIT License.
The models and point clouds are each licensed from a third party and are not part of this license! You should NOT distribute or use these models against their license. A list of the used model licenses and source is at models/README.md.