MLX, or MeshLabXML, is a Python (2.7 or 3) scripting interface to MeshLab, the open source system for processing and editing 3D triangular meshes.
Under the hood, MLX generates XML filter scripts that can then be executed headless with the meshlabserver executable or run in the MeshLab GUI. It can also parse some of MeshLab's output, such as the results of the measure_geometry and measure_topology functions.
MLX is named after the .mlx file extension for MeshLab script files, however the name was already taken on PyPi (for an unrelated machine learning library), so it is formally registered under the longer name of MeshLabXML.
MLX can be installed via PyPI and pip:
pip install meshlabxml
The released PyPI version may lag behind this git repository somewhat, so install from git if you want the latest and greatest. MLX may also be installed and run in other Python environments, such as Blender. Note that Blender does not come with pip by default, however it can be easily installed using get-pip.
Platforms: MLX should work anywhere that MeshLab will run, including Windows, Mac & Linux, although it is only routinely tested on 64 bit Windows.
Python: MLX should work under Python 2.7 and 3.x, although it is only routinely tested on 64 bit >=3.5
MeshLab: MLX is known to work on MeshLab versions 1.34BETA (64 bit Windows only) and 2016.12. As of the time of this writing, not all functions have been tested with 2016.12 yet, so please open an issue if you find a bug.
MLX version numbers: PyPI releases are numbered by the year and month of release, e.g. 2017.9. A letter may be added on the end ("a", "b", "c", etc) if there is more than one release in a month.
MLX contains a fairly large subset of the filters available in MeshLab. Additional filters will be added over time, generally on an "as I need them" basis. If you need a filter that is not yet incorporated, please open an issue.
Many of the functions below are a direct implementation of a MeshLab filter. Others are created from a combination of other functions, or implement new functionality using the muparser function filters.
Documentation for most filters is available by using "help" within a Python shell, although there are many that still need to be documented. In addition, in many cases the documentation is taken directly from the MeshLab filter, which is not always sufficient to understand the function if you are not already familiar with how it works.
mlx - functions to create and run scripts, determine inputs & outputs, etc.
- FilterScript - Main class to create scripts
- create_mlp
- find_texture_files
- default_output_mask
- run
mlx.create - functions that create a new mesh
- grid
- cube
- cube_hires
- cube_open_hires
- cylinder
- cylinder_open_hires
- tube_hires
- icosphere
- half_sphere_hires
- sphere_cap
- plane_hires_edges
- annulus
- annulus_hires
- torus
mlx.transform - functions that transform, deform or morph mesh geometry
- translate
- translate2
- rotate
- rotate2
- scale
- scale2
- freeze_matrix
- function
- function_cyl_co
- wrap2cylinder
- wrap2sphere
- emboss_sphere
- bend
- deform2curve
mlx.select - functions that work with selections
- all
- none
- invert
- border
- grow
- shrink
- self_intersecting_face
- nonmanifold_vert
- nonmanifold_edge
- small_parts
- vert_quality
- face_function
- vert_function
- cylindrical_vert
- spherical_vert
mlx.delete - functions that delete faces and/or vertices
- nonmanifold_vert
- nonmanifold_edge
- small_parts
- selected
- faces_from_nonmanifold_edges
- unreferenced_vert
- duplicate_faces
- duplicate_verts
- zero_area_face
mlx.clean - functions to clean and repair a mesh
- merge_vert
- close_holes
- split_vert_on_nonmanifold_face
- fix_folded_face
- snap_mismatched_borders
mlx.layers - functions that work with mesh layers
- join
- delete
- rename
- change
- duplicate
- split_parts
mlx.normals - functions that work with normals
- reorient
- flip
- fix
- point_sets
mlx.remesh - remeshing functions
- simplify
- uniform_resampling
- hull
- surface_poisson
- surface_poisson_screened
- curvature_flipping
- voronoi
mlx.sampling - sampling functions
- hausdorff_distance
- poisson_disk
- mesh_element
- clustered_vert
mlx.smooth - smoothing functions
- laplacian
- hc_laplacian
- taubin
- twostep
- depth
mlx.subdivide - subdivision functions
- loop
- ls3loop
- midpoint
- butterfly
- catmull_clark
mlx.texture - functions that work with textures and UV mapping (parameterization)
- flat_plane
- per_triangle
- voronoi
- isometric
- isometric_build_atlased_mesh
- isometric_save
- isometric_load
- isometric_transfer
- isometric_remesh
- set_texture
- project_rasters
- param_texture_from_rasters
- param_from_rasters
mlx.transfer - functions to transfer attributes
- tex2vc
- vc2tex
- fc2vc
- vc2fc
- mesh2fc
- vert_attr_2_meshes
- vert_attr2tex_2_meshes
- tex2vc_2_meshes
mlx.compute - functions that measure or perform a computation
- section
- measure_geometry
- measure_topology
- parse_geometry
- parse_topology
mlx.vert_color - functions that work with vertex colors
- function
- voronoi
- cyclic_rainbow
mlx.mp_func - functions to work with muparser filter functions, this is mostly a vector math library
- muparser_ref
- v_cross
- v_dot
- v_add
- v_subtract
- v_multiply
- v_length
- v_normalize
- torus_knot
- torus_knot_bbox
- vert_attr
- face_attr
- vq_function
- fq_function
mlx.files - functions that operate directly on files, usually to measure them
- measure_aabb
- measure_section
- measure_geometry
- measure_topology
- measure_all
- measure_dimension
For production MLX can run completely headless, however while developing new scripts some visual feedback may be helpful.
MLX does not have an integrated GUI like OpenSCAD or Blender, however you can simulate one by arranging several programs into a useful layout, such as shown below. This can be accomplished pretty easily in modern versions of Windows using the Windows key and the arrow keys.
Generally you will want a text editor, a console and of course MeshLab itself. The general script development workflow may consist of the following steps
- Write & edit script in text editor
- Run script in console and view meshlabserver output
- Load output in MeshLab. Use the "reload" button to reload the mesh after any changes are made.
- Repeat ad nauseam
Some simple examples are shown below. These assume that the meshlabserver executable is in your path. If it is not already in your path, you can add it to your path in your script using something similar to the following:
import os
meshlabserver_path = 'C:\\Program Files\\VCG\\MeshLab'
os.environ['PATH'] = meshlabserver_path + os.pathsep + os.environ['PATH']
Example #1: Create an orange cube and apply some transformations
import meshlabxml as mlx
orange_cube = mlx.FilterScript(file_out='orange_cube.ply', ml_version='2016.12')
mlx.create.cube(orange_cube, size=[3.0, 4.0, 5.0], center=True, color='orange')
mlx.transform.rotate(orange_cube, axis='x', angle=45)
mlx.transform.rotate(orange_cube, axis='y', angle=45)
mlx.transform.translate(orange_cube, value=[0, 5.0, 0])
orange_cube.run_script()
Output:
Example #2: Measure the built-in Stanford Bunny test model and print the results
import meshlabxml as mlx
aabb, geometry, topology = mlx.files.measure_all('bunny', ml_version='2016.12')
Output:
max = [103.817589, 87.032661, 191.203903]
diagonal = 310.4472554416525
size = [193.95133199999998, 149.001499, 191.203903]
center = [6.841923000000001, 12.531911500000003, 95.6019515]
min = [-90.133743, -61.968838, 0.0]
volume_cm3 = 1486.804
inertia_tensor = [[3170550016.0, -46370464.0, 987163904.0], [-46370464.0, 5072923136.0, -58690096.0], [987163904.0, -58690096.0, 3817212928.0]]
area_cm2 = 901.506875
center_of_mass = [1.747876, -2.226919, 64.788971]
total_edge_length = 179819.484375
volume_mm3 = 1486804.0
principal_axes = [[0.809736, 0.001188, -0.586793], [0.581329, 0.134539, 0.802468], [-0.0799, 0.990908, -0.108251]]
total_edge_length_incl_faux = 179819.484375
area_mm2 = 90150.6875
axis_momenta = [2455110656.0, 4522500608.0, 5083073024.0]
genus = 0
manifold = True
non_manifold_E = 0
vert_num = 32285
boundry_edge_num = 0
hole_num = 0
unref_vert_num = 0
edge_num = 96849
face_num = 64566
non_manifold_V = 0
part_num = 1
Check out the "examples" directory for more complex examples.
The MLX logo is a rainbow colored quatrefoil torus knot with Voronoi meshing. This model is created entirely using MLX; it's source code is included in the "examples" directory. This is a moderately complex script that should give you an idea of some of the things that MLX can do; it makes heavy use of the powerful muparser functions.
- MeshLabServer can be a bit unstable, especially on certain filters such as mlx.remesh.uniform_resampling. If you have many filters to run, it is better to break the project up into smaller scripts and run problematic filters or sequences independently.
- It is not currently possible to measure a mesh and use the results in the same script; you will need to measure the mesh and input the results into another instance. For example, if you want to simplify a mesh based on a percentage of the number of faces, you would first need to measure the number of faces with mlx.compute.measure_topology, then input the results into mlx.remesh.simplify.
MLX is still under heavy development and the API is not yet considered stable. Still, it is already quite useful, and is used for production purposes within our small company.
MLX is released under LGPL version 2.1
Example code is released into the public domain, except the quatrefoil logo, which is LGPL.
Any included models (such as the Stanford Bunny) are released under their own licenses.
Much of the documentation for the filter functions is taken directly from MeshLab, and is under the MeshLab license GPL version 3