Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Fast Tetrahedral Meshing in the Wild

Yixin Hu, Teseo Schneider, Bolun Wang, Denis Zorin, Daniele Panozzo. ACM Transactions on Graphics (SIGGRAPH 2020)

author = {Hu, Yixin and Schneider, Teseo and Wang, Bolun and Zorin, Denis and Panozzo, Daniele},
title = {Fast Tetrahedral Meshing in the Wild},
year = {2020},
issue_date = {July 2020},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {39},
number = {4},
issn = {0730-0301},
url = {},
doi = {10.1145/3386569.3392385},
journal = {ACM Trans. Graph.},
month = jul,
articleno = {117},
numpages = {18},
keywords = {mesh generation, robust geometry processing, tetrahedral meshing}

Important Tips

💡💡💡 If you are interested in the algorithm details, please refer to our paper first. We provide plenty of examples and statistics in the paper.

💡💡💡 Check our license first.

💡💡💡 Our algorithm is robust both in theory and in practice. If you do find fTetWild crash (on your laptop), please test it (on cluster) with more resource given. The most complex model I tested requires >32GB memory.

💡💡💡 The orientation of input faces is as important as the position of their vertices. Our algorithm is faithful to the input face position and orientation check Tetwild


Here is pre-generated tetmeshes and the extracted surface meshes for research-purpose usage. Please kindly cite our paper when using our pre-generated data.

Installation via CMake

Our code was originally developed on MacOS and has been tested on Linux and Windows. We provide the commands for installing fTetWild in MacOS:

  • Clone the repository into your local machine:
git clone
  • Compile the code using cmake (default in Release mode):
cd fTetWild
mkdir build
cd build
cmake ..

You may need to install gmp before compiling the code. You can install it via

brew install gmp
  • Package manager on Unix:
sudo apt-get install gmp
conda install -c conda-forge mpir

Note Windows The executable needs that the file mpir.dll is in the same directiory of FloatTetwild_bin.exe. Once you compliled the code, copy mpir.dll (e.g., <conda_dir>\Library\bin) to the directoy containing FloatTetwild_bin.exe.

Note if cmake cannot find gmp you need to export the envirnement variable GMP_INC and GMP_LIB to the folder where you installed (e.g., <conda_dir>\Library\include for GMP_INC and <conda_dir>\Library\lib for GMP_LIB).

  • Check the installation:
./FloatTetwild_bin --help

This command should show a list of fTetWild parameters.


Input/output Format

The inputs of our software are triangle surface meshes in .off/.obj/.stl/.ply format.

We support .mesh/.msh format output. The default output format is .msh with minimum dihedral angle recorded as element scalar field, which can be visualized by software Gmsh. You can use PyMesh::MshLoader and PyMesh::MshSaver in pymesh/ for read and write .msh meshes.


Our software is quite easy to use. Basically, users only need to provide a surface triangle mesh as input and our mesher would output a tetrahedral mesh by using default settings. If you want to customize your own tetmeshes, we also provide some options.

  • Envelope of size epsilon

Using smaller envelope preserves features better but also takes longer time. The default value of epsilon is b/1000, where b is the length of the diagonal of the bounding box.

  • Ideal edge length

Using smaller ideal edge length gives a denser mesh but also takes longer time. The default ideal edge length is b/20

  • Filtering energy

Our mesher stops optimizing the mesh when maximum energy is smaller than filtering energy. Thus, larger filtering energy means less optimization and sooner stopping. If you do not care about quality, then give a larger filtering energy would let you get the result earlier. The energy we used here is conformal AMIPS whose range is from 3 to +inf. The default filtering energy is 10.

💡 We suggest not to set filtering energy smaller than 8 for complex input.

  • Maximum number of optimization passes

Our mesher stops optimizing the mesh when the maximum number of passes is reached. The default number is 80.

  • Sizing field Users can provide a background tetmesh in .msh format with vertex scalar field values stored. The scalar field values is used for controlling edge length. The scalars inside an element of the background mesh are linearly interpolated.

💡 Here is an example including input surface mesh, background mesh and output tetmeshes with/without sizing control.

  • Smoothing open regions

Our method can fill gaps and holes but the tetmesh faces on those parts could be bumpy. We provide users an option to do Lapacian smoothing on those faces to get a smoother surface.

Command Line Switches

Our software supports usage via command line or via a C++ function wrapper. Here is an overview of all command line switches:

Usage: ./FloatTetwild_bin [OPTIONS]
  -h,--help                   Print this help message and exit
  -i,--input TEXT:FILE        Input surface mesh INPUT in .off/.obj/.stl/.ply format. (string, required)
  -o,--output TEXT            Output tetmesh OUTPUT in .msh format. (string, optional, default: input_file+postfix+'.msh')
  --tag TEXT:FILE             Tag input faces for Boolean operation.
  --op INT                    Boolean operation: 0: union, 1: intersection, 2: difference.
  -l,--lr FLOAT               ideal_edge_length = diag_of_bbox * L. (double, optional, default: 0.05)
  -e,--epsr FLOAT             epsilon = diag_of_bbox * EPS. (double, optional, default: 1e-3)
  --stop-energy FLOAT         Stop optimization when max energy is lower than this.
  --log TEXT                  Log info to given file.
  --level INT                 Log level (0 = most verbose, 6 = off).
  -q,--is-quiet               Mute console output. (optional)
  --skip-simplify             skip preprocessing.
  --no-binary                 export meshes as ascii
  --no-color                  don't export color
  --smooth-open-boundary      Smooth the open boundary.
  --manifold-surface          Force the output to be manifold.
  --coarsen                   Coarsen the output as much as possible.
  --csg TEXT:FILE             json file containg a csg tree
  --disable-filtering         Disable filtering out outside elements.
  --use-floodfill             Use flood-fill to extract interior volume.
  --use-general-wn            Use general winding number.
  --use-input-for-wn          Use input surface for winding number.
  --bg-mesh TEXT:FILE         Background mesh for sizing field (.msh file).
  --epsr-tags TEXT:FILE       List of envelope size for each input faces.
  --max-threads UINT          Maximum number of threads used


We used several useful libraries in our implement, testing, and rendering listed as follows. We would like to especially thank their authors for their great work and publishing the code.


Fast Tetrahedral Meshing in the Wild




No releases published


No packages published

Contributors 4