Home

Awesome

trimesh


Github Actions codecov Docker Image Version (latest by date) PyPI version

Trimesh is a pure Python 3.7+ library for loading and using triangular meshes with an emphasis on watertight surfaces. The goal of the library is to provide a full featured and well tested Trimesh object which allows for easy manipulation and analysis, in the style of the Polygon object in the Shapely library.

The API is mostly stable, but this should not be relied on and is not guaranteed: install a specific version if you plan on deploying something using trimesh.

Pull requests are appreciated and responded to promptly! If you'd like to contribute, here is an up to date list of potential enhancements although things not on that list are also welcome. Here's a quick development and contributing guide.

Basic Installation

Keeping trimesh easy to install is a core goal, thus the only hard dependency is numpy. Installing other packages adds functionality but is not required. For the easiest install with just numpy, pip can generally install trimesh cleanly on Windows, Linux, and OSX:

pip install trimesh

The minimal install can load many supported formats (STL, PLY, GLTF/GLB) into numpy arrays. More functionality is available when soft dependencies are installed. This includes things like convex hulls (scipy), graph operations (networkx), faster ray queries (embreex), vector path handling (shapely and rtree), XML formats like 3DXML/XAML/3MF (lxml), preview windows (pyglet), faster cache checks (xxhash), etc. To install trimesh with the soft dependencies that generally install cleanly on Linux, OSX, and Windows using pip:

pip install trimesh[easy]

Further information is available in the advanced installation documentation.

Quick Start

Here is an example of loading a mesh from file and colorizing its faces. Here is a nicely formatted ipython notebook version of this example. Also check out the cross section example.

import numpy as np
import trimesh

# attach to logger so trimesh messages will be printed to console
trimesh.util.attach_to_log()

# mesh objects can be created from existing faces and vertex data
mesh = trimesh.Trimesh(vertices=[[0, 0, 0], [0, 0, 1], [0, 1, 0]],
                       faces=[[0, 1, 2]])

# by default, Trimesh will do a light processing, which will
# remove any NaN values and merge vertices that share position
# if you want to not do this on load, you can pass `process=False`
mesh = trimesh.Trimesh(vertices=[[0, 0, 0], [0, 0, 1], [0, 1, 0]],
                       faces=[[0, 1, 2]],
                       process=False)

# some formats represent multiple meshes with multiple instances
# the loader tries to return the datatype which makes the most sense
# which will for scene-like files will return a `trimesh.Scene` object.
# if you *always* want a straight `trimesh.Trimesh` you can ask the
# loader to "force" the result into a mesh through concatenation
mesh = trimesh.load('models/CesiumMilkTruck.glb', force='mesh')

# mesh objects can be loaded from a file name or from a buffer
# you can pass any of the kwargs for the `Trimesh` constructor
# to `trimesh.load`, including `process=False` if you would like
# to preserve the original loaded data without merging vertices
# STL files will be a soup of disconnected triangles without
# merging vertices however and will not register as watertight
mesh = trimesh.load('../models/featuretype.STL')

# is the current mesh watertight?
mesh.is_watertight

# what's the euler number for the mesh?
mesh.euler_number

# the convex hull is another Trimesh object that is available as a property
# lets compare the volume of our mesh with the volume of its convex hull
print(mesh.volume / mesh.convex_hull.volume)

# since the mesh is watertight, it means there is a
# volumetric center of mass which we can set as the origin for our mesh
mesh.vertices -= mesh.center_mass

# what's the moment of inertia for the mesh?
mesh.moment_inertia

# if there are multiple bodies in the mesh we can split the mesh by
# connected components of face adjacency
# since this example mesh is a single watertight body we get a list of one mesh
mesh.split()

# facets are groups of coplanar adjacent faces
# set each facet to a random color
# colors are 8 bit RGBA by default (n, 4) np.uint8
for facet in mesh.facets:
    mesh.visual.face_colors[facet] = trimesh.visual.random_color()

# preview mesh in an opengl window if you installed pyglet and scipy with pip
mesh.show()

# transform method can be passed a (4, 4) matrix and will cleanly apply the transform
mesh.apply_transform(trimesh.transformations.random_rotation_matrix())

# axis aligned bounding box is available
mesh.bounding_box.extents

# a minimum volume oriented bounding box also available
# primitives are subclasses of Trimesh objects which automatically generate
# faces and vertices from data stored in the 'primitive' attribute
mesh.bounding_box_oriented.primitive.extents
mesh.bounding_box_oriented.primitive.transform

# show the mesh appended with its oriented bounding box
# the bounding box is a trimesh.primitives.Box object, which subclasses
# Trimesh and lazily evaluates to fill in vertices and faces when requested
# (press w in viewer to see triangles)
(mesh + mesh.bounding_box_oriented).show()

# bounding spheres and bounding cylinders of meshes are also
# available, and will be the minimum volume version of each
# except in certain degenerate cases, where they will be no worse
# than a least squares fit version of the primitive.
print(mesh.bounding_box_oriented.volume,
      mesh.bounding_cylinder.volume,
      mesh.bounding_sphere.volume)

Features

Viewer

Trimesh includes an optional pyglet based viewer for debugging and inspecting. In the mesh view window, opened with mesh.show(), the following commands can be used:

If called from inside a jupyter notebook, mesh.show() displays an in-line preview using three.js to display the mesh or scene. For more complete rendering (PBR, better lighting, shaders, better off-screen support, etc) pyrender is designed to interoperate with trimesh objects.

Projects Using Trimesh

You can check out the Github network for things using trimesh. A select few:

Which Mesh Format Should I Use?

Quick recommendation: GLB or PLY. Every time you replace OBJ with GLB an angel gets its wings.

If you want things like by-index faces, instancing, colors, textures, etc, GLB is a terrific choice. GLTF/GLB is an extremely well specified modern format that is easy and fast to parse: it has a JSON header describing data in a binary blob. It has a simple hierarchical scene graph, a great looking modern physically based material system, support in dozens-to-hundreds of libraries, and a John Carmack endorsment. Note that GLTF is a large specification, and trimesh only supports a subset of features: loading basic geometry is supported, NOT supported are fancier things like animations, skeletons, etc.

In the wild, STL is perhaps the most common format. STL files are extremely simple: it is basically just a list of triangles. They are robust and are a good choice for basic geometry. Binary PLY files are a good step up, as they support indexed faces and colors.

Wavefront OBJ is also pretty common: unfortunately OBJ doesn't have a widely accepted specification so every importer and exporter implements things slightly differently, making it tough to support. It also allows unfortunate things like arbitrary sized polygons, has a face representation which is easy to mess up, references other files for materials and textures, arbitrarily interleaves data, and is slow to parse. Give GLB or PLY a try as an alternative!

How can I cite this library?

A question that comes up pretty frequently is how to cite the library. A quick BibTex recommendation:

@software{trimesh,
	author = {{Dawson-Haggerty et al.}},
	title = {trimesh},
	url = {https://trimesh.org/},
	version = {3.2.0},
	date = {2019-12-8},
}

Containers

If you want to deploy something in a container that uses trimesh automated debian:slim-bullseye based builds with trimesh and most dependencies are available on Docker Hub with image tags for latest, git short hash for the commit in main (i.e. trimesh/trimesh:0c1298d), and version (i.e. trimesh/trimesh:3.5.27):

docker pull trimesh/trimesh

Here's an example of how to render meshes using LLVMpipe and XVFB inside a container.