Install

pip install ncollpyde

Pre-built wheels are available for Linux, MacOS, and Windows. If you have a stable rust compiler, you should also be able to install from source.

Features

• Checking whether points are inside a volume defined by a triangular mesh

• Checking the intersection of line segments with the mesh

• Get the (signed) distance from points to the boundary of a mesh

Usage

This library implements most of its functionality through the Volume class, instantiated from an array of vertices, and an array of triangles as indices into the vertex array.

# get an array of vertices and triangles which refer to those points
import meshio
mesh = meshio.read("meshes/teapot.stl")

# use this library
from ncollpyde import Volume

volume = Volume(mesh.points, mesh.cells_dict["triangle"])

# or, for convenience
volume = Volume.from_meshio(mesh)

# containment checks: singular and multiple
assert [-2.30, -4.15,  1.90] in volume
assert np.array_equal(
    volume.contains(
        [
            [-2.30, -4.15, 1.90],
            [-0.35, -0.51, 7.61],
        ]
    ),
    [True, False]
)

# line segment intersection
seg_idxs, intersections, is_backface = volume.intersections(
    [[-10, -10, -10], [0, 0, 3], [10, 10, 10]],
    [[0, 0, 3], [10, 10, 10], [20, 20, 20]],
)
assert np.array_equal(seg_idxs, [0, 1])  # idx 2 does not intersect
assert np.array_equal(seg_idxs, [0, 1])
assert np.allclose(
    intersections,
    [
        [-2.23347309, -2.23347309, 0.09648498],
        [ 3.36591285, 3.36591285, 5.356139],
    ],
)
assert np.array_equal(
    is_backface,
    [False, True],
)

# distance from boundary (negative means internal)
assert np.array_equal(
    volume.distance([[10, 10, 10], [0, 0, 3]]),
    [10.08592464, -2.99951118],
)

See the API docs for more advanced usage.

Known issues

• Performance gains for multi-threaded queries are underwhelming, especially for ray intersections: see this issue

• Very rare false positives for containment

  • Due to a bug in the underlying library

  • Only happens when the point is outside the mesh and fires a ray which touches a single edge or vertex of the mesh.

  • Also affects is_backface result for ray intersection checks

• manylinux-compatible wheels are built on CI but not necessarily in your local environment. Always allow CI to deploy the wheels.

• If you are installing from a source distribution rather than a wheel, you need a compatible rust toolchain

• Meshes with >= ~4.3bn vertices are not supported, as the underlying library uses u32 to address them. This is probably not a problem at time of writing; such a mesh would take up hundreds of GB of RAM to operate on.

ncollpyde v0.11 was the last to support meshio < 4.0.

Acknowledgements

Thanks to top users Philipp Schlegel (check out navis!) and Nik Drummond for their help in debugging and expanding ncollpyde ‘s functionality.

Thanks also to pyo3/ maturin developers @konstin and @messense for taking an interest in the project and helping along the way.