Skip to content

pyvista/fast-simplification

Repository files navigation

Python Fast-Quadric-Mesh-Simplification Wrapper

This is a python wrapping of the Fast-Quadric-Mesh-Simplification Library. Having arrived at the same problem as the original author, but needing a Python library, this project seeks to extend the work of the original library while adding integration to Python and the PyVista project.

For the full documentation visit: https://pyvista.github.io/fast-simplification/

https://github.com/pyvista/fast-simplification/raw/main/doc/images/simplify_demo.png

Installation

Fast Simplification can be installed from PyPI using pip on Python >= 3.7:

pip install fast-simplification

See the Contributing for more details regarding development or if the installation through pip doesn't work out.

Basic Usage

The basic interface is quite straightforward and can work directly with arrays of points and triangles:

points = [[ 0.5, -0.5, 0.0],
          [ 0.0, -0.5, 0.0],
          [-0.5, -0.5, 0.0],
          [ 0.5,  0.0, 0.0],
          [ 0.0,  0.0, 0.0],
          [-0.5,  0.0, 0.0],
          [ 0.5,  0.5, 0.0],
          [ 0.0,  0.5, 0.0],
          [-0.5,  0.5, 0.0]]

faces = [[0, 1, 3],
         [4, 3, 1],
         [1, 2, 4],
         [5, 4, 2],
         [3, 4, 6],
         [7, 6, 4],
         [4, 5, 7],
         [8, 7, 5]]

points_out, faces_out = fast_simplification.simplify(points, faces, 0.5)

Advanced Usage

This library supports direct integration with VTK through PyVista to provide a simplistic interface to the library. As this library provides a 4-5x improvement to the VTK decimation algorithms.

>>> from pyvista import examples
>>> mesh = examples.download_nefertiti()
>>> out = fast_simplification.simplify_mesh(mesh, target_reduction=0.9)

Compare with built-in VTK/PyVista methods:

>>> fas_sim = fast_simplification.simplify_mesh(mesh, target_reduction=0.9)
>>> dec_std = mesh.decimate(0.9)  # vtkQuadricDecimation
>>> dec_pro = mesh.decimate_pro(0.9)  # vtkDecimatePro

>>> pv.set_plot_theme('document')
>>> pl = pv.Plotter(shape=(2, 2), window_size=(1000, 1000))
>>> pl.add_text('Original', 'upper_right', color='w')
>>> pl.add_mesh(mesh, show_edges=True)
>>> pl.camera_position = cpos

>>> pl.subplot(0, 1)
>>> pl.add_text(
...    'Fast-Quadric-Mesh-Simplification\n~2.2 seconds', 'upper_right', color='w'
... )
>>> pl.add_mesh(fas_sim, show_edges=True)
>>> pl.camera_position = cpos

>>> pl.subplot(1, 0)
>>> pl.add_mesh(dec_std, show_edges=True)
>>> pl.add_text(
...    'vtkQuadricDecimation\n~9.5 seconds', 'upper_right', color='w'
... )
>>> pl.camera_position = cpos

>>> pl.subplot(1, 1)
>>> pl.add_mesh(dec_pro, show_edges=True)
>>> pl.add_text(
...    'vtkDecimatePro\n11.4~ seconds', 'upper_right', color='w'
... )
>>> pl.camera_position = cpos
>>> pl.show()

Comparison to other libraries

The pyfqmr library wraps the same header file as this library and has similar capabilities. In this library, the decision was made to write the Cython layer on top of an additional C++ layer rather than directly interfacing with wrapper from Cython. This results in a mild performance improvement.

Reusing the example above:

Set up a timing function.

>>> import pyfqmr
>>> vertices = mesh.points
>>> faces = mesh.faces.reshape(-1, 4)[:, 1:]
>>> def time_pyfqmr():
...     mesh_simplifier = pyfqmr.Simplify()
...     mesh_simplifier.setMesh(vertices, faces)
...     mesh_simplifier.simplify_mesh(
...         target_count=out.n_faces, aggressiveness=7, verbose=0
...     )
...     vertices_out, faces_out, normals_out = mesh_simplifier.getMesh()
...     return vertices_out, faces_out, normals_out

Now, time it and compare with the non-VTK API of this library:

>>> timeit time_pyfqmr()
2.75 s ± 5.35 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

>>> timeit vout, fout = fast_simplification.simplify(vertices, faces, 0.9)
2.05 s ± 3.18 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Additionally, the fast-simplification library has direct plugins to the pyvista library, making it easy to read and write meshes:

>>> import pyvista
>>> import fast_simplification
>>> mesh = pyvista.read('my_mesh.stl')
>>> simple = fast_simplification.simplify_mesh(mesh)
>>> simple.save('my_simple_mesh.stl')

Since both libraries are based on the same core C++ code, feel free to use whichever gives you the best performance and interoperability.

Replay decimation functionality

This library also provides an interface to keep track of the successive collapses that occur during the decimation process and to replay the decimation process. This can be useful for different applications, such as:

  • applying the same decimation to a collection of meshes that share the same topology
  • computing a correspondence map between the vertices of the original mesh and the vertices of the decimated mesh, to transfer field data from one to the other for example
  • replaying the decimation process with a smaller target reduction than the original one, faster than decimating the original mesh with the smaller target reduction

To use this functionality, you need to set the return_collapses parameter to True when calling simplify. This will return the successive collapses of the decimation process in addition to points and faces.

>>> import fast_simplification
>>> import pyvista
>>> mesh = pyvista.Sphere()
>>> points, faces = mesh.points, mesh.faces.reshape(-1, 4)[:, 1:]
>>> points_out, faces_out, collapses = fast_simplification.simplify(points, faces, 0.9, return_collapses=True)

Now you can call replay_simplification to replay the decimation process and obtain the mapping between the vertices of the original mesh and the vertices of the decimated mesh.

>>> points_out, faces_out, indice_mapping = fast_simplification.replay_simplification(points, faces, collapses)
>>> i = 3
>>> print(f'Vertex {i} of the original mesh is mapped to {indice_mapping[i]} of the decimated mesh')

You can also use the replay_simplification function to replay the decimation process with a smaller target reduction than the original one. This is faster than decimating the original mesh with the smaller target reduction. To do so, you need to pass a subset of the collapses to the replay_simplification function. For example, to replay the decimation process with a target reduction of 50% the initial rate, you can run:

>>> import numpy as np
>>> collapses_half = collapses[:int(0.5 * len(collapses))]
>>> points_out, faces_out, indice_mapping = fast_simplification.replay_simplification(points, faces, collapses_half)

If you have a collection of meshes that share the same topology, you can apply the same decimation to all of them by calling replay_simplification with the same collapses for each mesh. This ensure that the decimated meshes will share the same topology.

>>> import numpy as np
>>> # Assume that you have a collection of meshes stored in a list meshes
>>> _, _, collapses = fast_simplification.simplify(meshes[0].points, meshes[0].faces,
...                                                0.9, return_collapses=True)
>>> decimated_meshes = []
>>> for mesh in meshes:
...     points_out, faces_out, _ = fast_simplification.replay_simplification(mesh.points, mesh.faces, collapses)
...     decimated_meshes.append(pyvista.PolyData(points_out, faces_out))

Contributing

Contribute to this repository by forking this repository and installing in development mode with:

git clone https://github.com/<USERNAME>/fast-simplification
pip install -e .
pip install -r requirements_test.txt

You can then add your feature or commit your bug fix and then run your unit testing with:

pytest

Unit testing will automatically enforce minimum code coverage standards.

Next, to ensure your code meets minimum code styling standards, run:

pip install pre-commit
pre-commit run --all-files

Finally, create a pull request from your fork and I'll be sure to review it.