Turn a volume into vessels
Requirements:
| Supported Compiler | Version >= |
|---|---|
| GCC | 8.1 |
| Clang | 7 |
| Apple Xcode | 11 |
| Required Libraries / Packages | Version >= |
|---|---|
| glm | 0.9.7 |
| OpenVDB | 6.1.0 |
| qmake1 | - |
1A full Qt install not required, only qmake.
To build, create a build directory somewhere, cd into it, and run qmake <path-to-vasc.pro>. You should then be able to run make.
As an example, look at the bunny case directory. Inside is a source wavefront object and a control file.
The control file has the following options:
| Option | Description |
|---|---|
mesh |
Path to the wavefront object to consume. |
voxel_size |
Size of voxels in mesh coordinate space. |
output |
Name of the output vascular mesh. |
position_randomness |
Vessel position randomness. |
prune |
Number of rounds of vessel leaves to prune. |
prune_flow |
Vessel sizes less than this value will be pruned. |
dump_voxels |
Write voxels to case directory, will appear as a csv. |
To start the run, pass the control file as the only argument to the vascularize executable.
To build the vascularized structure, the following procedure is used.
Input mesh is read in, and a voxel grid is created with a resolution based on voxel_size.
Each voxel is then given a value based on the inverse distance to the edge of the mesh, i.e. edge voxels will have a high value, center voxels will have a low value. Randomness is added.
A flow graph is created where high valued voxels are connected to low value voxels. A minimum spanning tree is built from this graph; weights for edges are the delta in voxel value. A root is selected (at this time, just picks the lowest voxel), and flow is computed from that point. Flow, in this case, is just the count of all downstream nodes.
The graph is then turned into a mesh. We first prune parts of the graph, leaves of the graph, and areas where the flow is too small. At this point the nodes of the graph have a position that is regularized on a grid, so positions are perturbed by randomness, and then the positions are relaxed to smooth out hard edges. Each edge is turned into a tube, and emitted into another voxel grid, which is then isosurfaced, and written to disk.