https://github.com/pdal/python
PDAL's Python Support
https://github.com/pdal/python
Last synced: about 1 year ago
JSON representation
PDAL's Python Support
- Host: GitHub
- URL: https://github.com/pdal/python
- Owner: PDAL
- License: other
- Created: 2018-03-14T15:13:38.000Z (over 8 years ago)
- Default Branch: main
- Last Pushed: 2024-12-17T20:51:10.000Z (over 1 year ago)
- Last Synced: 2025-04-01T08:42:09.124Z (over 1 year ago)
- Language: C++
- Homepage:
- Size: 1.27 MB
- Stars: 125
- Watchers: 12
- Forks: 37
- Open Issues: 9
-
Metadata Files:
- Readme: README.rst
- Changelog: CHANGES.txt
- License: LICENSE.txt
Awesome Lists containing this project
README
================================================================================
PDAL
================================================================================
PDAL Python support allows you to process data with PDAL into `Numpy`_ arrays.
It provides a PDAL extension module to control Python interaction with PDAL.
Additionally, you can use it to fetch `schema`_ and `metadata`_ from PDAL operations.
Installation
--------------------------------------------------------------------------------
**Note** The PDAL Python bindings require the PDAL base library installed. Source code can be found at https://pdal.io and `GitHub `__.
PyPI
................................................................................
PDAL Python support is installable via PyPI:
.. code-block::
pip install PDAL
Developers can control many settings including debug builds and where the libraries are installed
using `scikit-build-core `_ settings:
.. code-block::
python -m pip install \
-Cbuild-dir=build \
-e \
. \
--config-settings=cmake.build-type="Debug" \
-vv \
--no-deps \
--no-build-isolation
GitHub
................................................................................
The repository for PDAL's Python extension is available at https://github.com/PDAL/python
Python support released independently from PDAL itself as of PDAL 1.7.
Usage
--------------------------------------------------------------------------------
Simple
................................................................................
Given the following pipeline, which simply reads an `ASPRS LAS`_ file and
sorts it by the ``X`` dimension:
.. _`ASPRS LAS`: https://www.asprs.org/committee-general/laser-las-file-format-exchange-activities.html
.. code-block:: python
json = """
{
"pipeline": [
"1.2-with-color.las",
{
"type": "filters.sort",
"dimension": "X"
}
]
}"""
import pdal
pipeline = pdal.Pipeline(json)
count = pipeline.execute()
arrays = pipeline.arrays
metadata = pipeline.metadata
log = pipeline.log
Programmatic Pipeline Construction
................................................................................
The previous example specified the pipeline as a JSON string. Alternatively, a
pipeline can be constructed by creating ``Stage`` instances and piping them
together. For example, the previous pipeline can be specified as:
.. code-block:: python
pipeline = pdal.Reader("1.2-with-color.las") | pdal.Filter.sort(dimension="X")
Stage Objects
=============
- A stage is an instance of ``pdal.Reader``, ``pdal.Filter`` or ``pdal.Writer``.
- A stage can be instantiated by passing as keyword arguments the options
applicable to the respective PDAL stage. For more on PDAL stages and their
options, check the PDAL documentation on `Stage Objects `__.
- The ``filename`` option of ``Readers`` and ``Writers`` as well as the ``type``
option of ``Filters`` can be passed positionally as the first argument.
- The ``inputs`` option specifies a sequence of stages to be set as input to the
current stage. Each input can be either the string tag of another stage, or
the ``Stage`` instance itself.
- The ``Reader``, ``Filter`` and ``Writer`` classes come with static methods for
all the respective PDAL drivers. For example, ``pdal.Filter.head()`` is a
shortcut for ``pdal.Filter(type="filters.head")``. These methods are
auto-generated by introspecting ``pdal`` and the available options are
included in each method's docstring:
.. code-block::
>>> help(pdal.Filter.head)
Help on function head in module pdal.pipeline:
head(**kwargs)
Return N points from beginning of the point cloud.
user_data: User JSON
log: Debug output filename
option_file: File from which to read additional options
where: Expression describing points to be passed to this filter
where_merge='auto': If 'where' option is set, describes how skipped points should be merged with kept points in standard mode.
count='10': Number of points to return from beginning. If 'invert' is true, number of points to drop from the beginning.
invert='false': If true, 'count' specifies the number of points to skip from the beginning.
Pipeline Objects
================
A ``pdal.Pipeline`` instance can be created from:
- a JSON string: ``Pipeline(json_string)``
- a sequence of ``Stage`` instances: ``Pipeline([stage1, stage2])``
- a single ``Stage`` with the ``Stage.pipeline`` method: ``stage.pipeline()``
- nothing: ``Pipeline()`` creates a pipeline with no stages.
- joining ``Stage`` and/or other ``Pipeline`` instances together with the pipe
operator (``|``):
- ``stage1 | stage2``
- ``stage1 | pipeline1``
- ``pipeline1 | stage1``
- ``pipeline1 | pipeline2``
Every application of the pipe operator creates a new ``Pipeline`` instance. To
update an existing ``Pipeline`` use the respective in-place pipe operator (``|=``):
.. code-block:: python
# update pipeline in-place
pipeline = pdal.Pipeline()
pipeline |= stage
pipeline |= pipeline2
Reading using Numpy Arrays
................................................................................
The following more complex scenario demonstrates the full cycling between
PDAL and Python:
* Read a small testfile from GitHub into a Numpy array
* Filters the array with Numpy for Intensity
* Pass the filtered array to PDAL to be filtered again
* Write the final filtered array to a LAS file and a TileDB_ array
via the `TileDB-PDAL integration`_ using the `TileDB writer plugin`_
.. code-block:: python
import pdal
data = "https://github.com/PDAL/PDAL/blob/master/test/data/las/1.2-with-color.las?raw=true"
pipeline = pdal.Reader.las(filename=data).pipeline()
print(pipeline.execute()) # 1065 points
# Get the data from the first array
# [array([(637012.24, 849028.31, 431.66, 143, 1,
# 1, 1, 0, 1, -9., 132, 7326, 245380.78254963, 68, 77, 88),
# dtype=[('X', ' 30]
print(len(intensity)) # 704 points
# Now use pdal to clamp points that have intensity 100 <= v < 300
pipeline = pdal.Filter.expression(expression="Intensity >= 100 && Intensity < 300").pipeline(intensity)
print(pipeline.execute()) # 387 points
clamped = pipeline.arrays[0]
# Write our intensity data to a LAS file and a TileDB array. For TileDB it is
# recommended to use Hilbert ordering by default with geospatial point cloud data,
# which requires specifying a domain extent. This can be determined automatically
# from a stats filter that computes statistics about each dimension (min, max, etc.).
pipeline = pdal.Writer.las(
filename="clamped.las",
offset_x="auto",
offset_y="auto",
offset_z="auto",
scale_x=0.01,
scale_y=0.01,
scale_z=0.01,
).pipeline(clamped)
pipeline |= pdal.Filter.stats() | pdal.Writer.tiledb(array_name="clamped")
print(pipeline.execute()) # 387 points
# Dump the TileDB array schema
import tiledb
with tiledb.open("clamped") as a:
print(a.schema)
Reading using Numpy Arrays as buffers (advanced)
................................................................................
It's also possible to treat the Numpy arrays passed to PDAL as buffers that are iteratively populated through
custom python functions during the execution of the pipeline.
This may be useful in cases where you want the reading of the input data to be handled in a streamable fashion,
like for example:
* When the total Numpy array data wouldn't fit into memory.
* To initiate execution of a streamable PDAL pipeline while the input data is still being read.
To enable this mode, you just need to include the python populate function along with each corresponding Numpy array.
.. code-block:: python
# Numpy array to be used as buffer
in_buffer = np.zeros(max_chunk_size, dtype=[("X", float), ("Y", float), ("Z", float)])
# The function to populate the buffer iteratively
def load_next_chunk() -> int:
"""
Function called by PDAL before reading the data from the buffer.
IMPORTANT: must return the total number of items to be read from the buffer.
The Pipeline execution will keep calling this function in a loop until 0 is returned.
"""
#
# Replace here with your code that populates the buffer and returns the number of elements to read
#
chunk_size = next_chunk.size
in_buffer[:chunk_size]["X"] = next_chunk[:]["X"]
in_buffer[:chunk_size]["Y"] = next_chunk[:]["Y"]
in_buffer[:chunk_size]["Z"] = next_chunk[:]["Z"]
return chunk_size
# Configure input array and handler during Pipeline initialization...
p = pdal.Pipeline(pipeline_json, arrays=[in_buffer], stream_handlers=[load_next_chunk])
# ...alternatively you can use the setter on an existing Pipeline
# p.inputs = [(in_buffer, load_next_chunk)]
The following snippet provides a simple example of how to use a Numpy array as buffer to support writing through PDAL
with total control over the maximum amount of memory to use.
.. raw:: html
Example: Streaming the read and write of a very large LAZ file with low memory footprint
.. code-block:: python
import numpy as np
import pdal
in_chunk_size = 10_000_000
in_pipeline = pdal.Reader.las(**{
"filename": "in_test.laz"
}).pipeline()
in_pipeline_it = in_pipeline.iterator(in_chunk_size).__iter__()
out_chunk_size = 50_000_000
out_file = "out_test.laz"
out_pipeline = pdal.Writer.las(
filename=out_file
).pipeline()
out_buffer = np.zeros(in_chunk_size, dtype=[("X", float), ("Y", float), ("Z", float)])
def load_next_chunk():
try:
next_chunk = next(in_pipeline_it)
except StopIteration:
# Stops the streaming
return 0
chunk_size = next_chunk.size
out_buffer[:chunk_size]["X"] = next_chunk[:]["X"]
out_buffer[:chunk_size]["Y"] = next_chunk[:]["Y"]
out_buffer[:chunk_size]["Z"] = next_chunk[:]["Z"]
print(f"Loaded next chunk -> {chunk_size}")
return chunk_size
out_pipeline.inputs = [(out_buffer, load_next_chunk)]
out_pipeline.loglevel = 20 # INFO
count = out_pipeline.execute_streaming(out_chunk_size)
print(f"\nWROTE - {count}")
.. raw:: html
Executing Streamable Pipelines
................................................................................
Streamable pipelines (pipelines that consist exclusively of streamable PDAL
stages) can be executed in streaming mode via ``Pipeline.iterator()``. This
returns an iterator object that yields Numpy arrays of up to ``chunk_size`` size
(default=10000) at a time.
.. code-block:: python
import pdal
pipeline = pdal.Reader("test/data/autzen-utm.las") | pdal.Filter.expression(expression="Intensity > 80 && Intensity < 120)")
for array in pipeline.iterator(chunk_size=500):
print(len(array))
# or to concatenate all arrays into one
# full_array = np.concatenate(list(pipeline))
``Pipeline.iterator()`` also takes an optional ``prefetch`` parameter (default=0)
to allow prefetching up to to this number of arrays in parallel and buffering
them until they are yielded to the caller.
If you just want to execute a streamable pipeline in streaming mode and don't
need to access the data points (typically when the pipeline has Writer stage(s)),
you can use the ``Pipeline.execute_streaming(chunk_size)`` method instead. This
is functionally equivalent to ``sum(map(len, pipeline.iterator(chunk_size)))``
but more efficient as it avoids allocating and filling any arrays in memory.
Accessing Mesh Data
................................................................................
Some PDAL stages (for instance ``filters.delaunay``) create TIN type mesh data.
This data can be accessed in Python using the ``Pipeline.meshes`` property, which returns a ``numpy.ndarray``
of shape (1,n) where n is the number of Triangles in the mesh.
If the PointView contains no mesh data, then n = 0.
Each Triangle is a tuple ``(A,B,C)`` where A, B and C are indices into the PointView identifying the point that is the vertex for the Triangle.
Meshio Integration
................................................................................
The meshes property provides the face data but is not easy to use as a mesh. Therefore, we have provided optional Integration
into the `Meshio `__ library.
The ``pdal.Pipeline`` class provides the ``get_meshio(idx: int) -> meshio.Mesh`` method. This
method creates a `Mesh` object from the `PointView` array and mesh properties.
.. note:: The meshio integration requires that meshio is installed (e.g. ``pip install meshio``). If it is not, then the method fails with an informative RuntimeError.
Simple use of the functionality could be as follows:
.. code-block:: python
import pdal
...
pl = pdal.Pipeline(pipeline)
pl.execute()
mesh = pl.get_meshio(0)
mesh.write('test.obj')
Advanced Mesh Use Case
................................................................................
USE-CASE : Take a LiDAR map, create a mesh from the ground points, split into tiles and store the tiles in PostGIS.
.. note:: Like ``Pipeline.arrays``, ``Pipeline.meshes`` returns a list of ``numpy.ndarray`` to provide for the case where the output from a Pipeline is multiple PointViews
(example using 1.2-with-color.las and not doing the ground classification for clarity)
.. code-block:: python
import pdal
import psycopg2
import io
pl = (
pdal.Reader(".../python/test/data/1.2-with-color.las")
| pdal.Filter.splitter(length=1000)
| pdal.Filter.delaunay()
)
pl.execute()
conn = psycopg(%CONNNECTION_STRING%)
buffer = io.StringIO
for idx in range(len(pl.meshes)):
m = pl.get_meshio(idx)
if m:
m.write(buffer, file_format = "wkt")
with conn.cursor() as curr:
curr.execute(
"INSERT INTO %table-name% (mesh) VALUES (ST_GeomFromEWKT(%(ewkt)s)",
{ "ewkt": buffer.getvalue()}
)
conn.commit()
conn.close()
buffer.close()
.. _`Numpy`: http://www.numpy.org/
.. _`schema`: http://www.pdal.io/dimensions.html
.. _`metadata`: http://www.pdal.io/development/metadata.html
.. _`TileDB`: https://tiledb.com/
.. _`TileDB-PDAL integration`: https://docs.tiledb.com/geospatial/pdal
.. _`TileDB writer plugin`: https://pdal.io/stages/writers.tiledb.html
.. image:: https://github.com/PDAL/python/workflows/Build/badge.svg
:target: https://github.com/PDAL/python/actions?query=workflow%3ABuild
Requirements
================================================================================
* PDAL 2.6+
* Python >=3.9
* Pybind11 (eg :code:`pip install pybind11[global]`)
* Numpy >= 1.22 (eg :code:`pip install numpy`)
* scikit-build-core (eg :code:`pip install scikit-build-core`)