priarobotics / voxsurf

A simple, easily hackable C++ surface voxelizer (STL=>voxels)

VoxSurf

A simple, easily hackable C++ surface voxelizer (STL=>voxels)

Takes as input a file 'model.stl' from the source directory. Outputs a voxel file named 'out.slab.vox' that can be read by MagicaVoxel https://ephtracy.github.io/ This is the 'slab' format as exported by MagicaVoxel, which is the format used by the Slab6 editor http://www.advsys.net/ken/download.htm.

This is meant as an introductory, easily modifiable implementation, and includes absoltutely zero optimization (e.g. SSE and multi-core would come to mind) nor any option such as performing a conservative voxelization. For other implementations, see links below.

The basic principle is to rasterize triangles using three 2D axis aligned grids, using integer arithmetic (fixed floating point) for robust triangle interior checks.

The code now supports filling the interior with voxels, with an optional voting scheme in case the input mesh has cracks (not strictly watertight). The scheme is quite simple and efficient. A bit is flipped every time a surface is contained in a voxel. After all surfaces are rasterized into voxels, if the bit is set the voxel is considered on the boundary, otherwise it is considered empty. Thus, only voxels crossed by odd number of surfaces are considered as belonging to the boundary (this is a form of winding number). The intervals in between boundary voxels are then filled. This is done for all three directions to allow for a voting scheme in case the input mesh has cracks.

Very simple, CPU only, no dependencies and surprisingly efficient despite the straightforward implementation. Higher resolutions could easily be reached by not storing the voxels as a dense 3D array (e.g. use blocking or an octree).

Here is a relatively large model voxelized at 1024^3 in ~1.5 seconds on a Core i5-3570, 3.4GHz. (Model: Ford engine block by Ford, rendered in MagicaVoxel viewer).

Compiling

Clone the main repo, then enter the directory and type:

git submodule init
git submodule update
cmake .
make

Tested with Viusal Studio 2017 and gcc 6.2.1

Links

• CUDA voxelization (GPU, fast) https://github.com/Forceflow/cuda_voxelizer
• Michael Schwarz and Hans-Peter Seidel paper on the topic http://research.michael-schwarz.com/publ/files/vox-siga10.pdf
• Header only voxelization in C https://github.com/karimnaaji/voxelizer

Credits

The included STL model is 3D knot by chylld