I used SciPy’s scipy.spatial.Voronoi to perform a 3D Voronoi partition, clipped the Voronoi regions (polyhedra) to a cube or sphere, and created 3D models of the Voronoi regions from the clipped vertices. Clone the repository, run clip_to_cube.py or clip_to_sphere.py, and press the [U] key on your keyboard. A force is applied to the 3D models, causing them to collapse. The parameters for the Voronoi partitioning and 3D models are managed in clipping_config.yaml. If you want to increase the number of Voronoi cuts (cut_points) or the number of times the 3D model’s faces are subdivided into triangles (max_depth), setting multi_processing to true will speed up the generation of the 3D models.

• Panda3D 1.10.16
• numpy 2.2.6
• scipy 1.16.2
• shapely 2.1.2
• PyYAML 6.0.3

• Python 3.13
• Windows11

git clone --recursive https://github.com/taKana671/Clipped3DVoronoi.git

# cube
python clip_to_cube.py

# sphere
python clip_to_sphere.py

After profiling with cProfile and refactoring the code, the time taken to run clip_to_cube.py or clip_to_sphere.py with the default settings in clipping_config.yaml is now around 4 seconds, which is nearly the same as when using multiple processes. Some NumPy methods (such as numpy.cross) were taking a significant amount of time when repeatedly called with small arrays like shape=(3,) as arguments. After changing the implementation to avoid using these NumPy methods, the speed improved.