The default PLY importer in Blender is quite slow when it comes to importing large mesh files. Blender 3.6 introduced a new C++-based PLY importer, which is marked "experimental", see below for comparison against this importer.
Most Python-based importers suffer from slow import because during import Python data structures are built up holding all geometry, vertex colors, etc. This simply takes quite a lot of time (and memory).
Fortunately, Python objects that support the buffer protocol can be
passed in certain places of the Blender Python API.
We can use this functionality to pass chunks of memory containing
vertex and face data, without having to build up Python data
structures. We use NumPy arrays in the readply
extension module
to easily pass the data directly to Blender.
Example usage:
$ blender [<scene.blend>] -P mesh_readply.py -- myfile.ply
Notes:
readply
module is not tied to Blender in any way and can
be used as a general PLY reader in Python.Below are some numbers when importing the Asian Dragon model [1] from The Stanford 3D Scanning Repository [2]. This 3D model consists of 3,609,600 vertices and 7,219,045 triangles, with the file being around 137 MB.
With Blender 3.6.5 and xyzrgb_dragon.ply
already in the filesystem cache:
# Native blender PLY importer (bpy.ops.import_mesh.ply())
$ blender -P test/blender_native_import.py -- xyzrgb_dragon.ply
...
Successfully imported '/home/melis/models/stanford/xyzrgb_dragon.ply' in 40.316 sec
Imported in 40.601s (legacy)
# New experimental blender PLY importer available in 3.6+ (bpy.ops.wm.ply_import())
$ blender -P test/blender_native_import_experimental.py -- xyzrgb_dragon.ply
...
PLY import of 'xyzrgb_dragon.ply' took 1471.2 ms
Imported in 1.744s (experimental)
# mesh_readply.py using readply extension module
$ blender -P mesh_readply.py -- xyzrgb_dragon.ply
...
PLY file read by readply() in 0.774s
Blender object+mesh created in 5.812s
Total import time 6.586s
I.e. in this test the mesh_readply.py
script (which uses the readply
module) loads the Dragon model 6.2x faster into Blender than
Blender's own legacy PLY import script, but 3.8x slower than the new
experimental PLY importer. The latter appears to be mostly due to a
much higher time to create the mesh from the imported data.
A setup.py
script is provided to build the extension, either under
regular Python or with Blender's included version of Python.
Note that for Blender usage it is advised to build and install the module using Blender's Python version, as that will take care of placing the module in the correct location.
[!NOTE] The official Blender binaries do not include the Python headers. So you still need a full Python installation somewhere to build the
readply
extension.
Run the setup.py
script with Blender's copy of
the Python interpreter. There should be a python3.10
executable in
your Blender directory. For example, for 3.6.5 on Linux the Python binary
is located at <blender-dir>/3.6/python/bin/python3.10
. Then run
$ <blender-dir>/3.6/python/bin/python3.10 setup.py install
If you get an error regarding the setuptools
module not being found,
then run .../python3.10 -m ensurepip
which should install the
pip
module, followed by installing the setuptools
module.
An alternative way is to run the setup script under Blender:
$ blender -b -P setup.py
There's at least two options:
# Build the module, then copy it to the top-level directory
$ python setup.py build_ext --inplace
or
# Build the module, then copy it to the default Python module location
# (which might be a system-wide directory)
$ python setup.py install
readply
extension module can be compiled for both Python 2.x and 3.x,
even though Blender uses Python 3.xPaul Melis (paul.melis@surf.nl), SURF Visualization team
The files under the rply/
directory are a copy of the RPly 1.1.4
source distribution (see http://w3.impa.br/~diego/software/rply/).
See the LICENSE
file in the root directory of this distribution,
which applies to all files except the ones in the rply/
directory.
See rply/LICENSE
for the the license of the RPly sources.