py4dgeo 1.0.0__cp314-cp314t-win_amd64.whl

py4dgeo/fallback.py

@@ -0,0 +1,159 @@
+"""Fallback implementations for C++ components of the M3C2 algorithms"""
+
+from py4dgeo.m3c2 import M3C2
+
+import numpy as np
+import _py4dgeo
+
+
+def radius_workingset_finder(params: _py4dgeo.WorkingSetFinderParameters) -> np.ndarray:
+    indices = params.epoch._radius_search(params.corepoint, params.radius)
+    return params.epoch._cloud[indices, :]
+
+
+def cylinder_workingset_finder(
+    params: _py4dgeo.WorkingSetFinderParameters,
+) -> np.ndarray:
+    # Cut the cylinder into N segments, perform radius searches around the
+    # segment midpoints and create the union of indices
+    N = 1
+    max_cylinder_length = params.max_distance
+    if max_cylinder_length >= params.radius:
+        N = np.ceil(max_cylinder_length / params.radius)
+    else:
+        max_cylinder_length = params.radius
+
+    r_cyl = np.sqrt(
+        params.radius * params.radius
+        + max_cylinder_length * max_cylinder_length / (N * N)
+    )
+
+    slabs = []
+    for i in range(int(N)):
+        # Find indices around slab midpoint
+        qp = (
+            params.corepoint[0, :]
+            + (2 * i - N + 1) / N * max_cylinder_length * params.cylinder_axis[0, :]
+        )
+        indices = params.epoch._radius_search(qp, r_cyl)
+
+        # Gather the points from the point cloud
+        superset = params.epoch._cloud[indices, :]
+
+        # Calculate distance from the axis and the plane perpendicular to the axis
+        to_corepoint = superset - qp
+        to_corepoint_plane = to_corepoint.dot(params.cylinder_axis[0, :])
+        to_axis2 = np.sum(
+            np.square(
+                to_corepoint
+                - np.multiply(
+                    to_corepoint_plane[:, np.newaxis], params.cylinder_axis[0, :]
+                )
+            ),
+            axis=1,
+        )
+
+        # Filter the points that are not within the slab
+        filtered = superset[
+            np.logical_and(
+                to_axis2 <= params.radius * params.radius,
+                np.abs(to_corepoint_plane) < max_cylinder_length / N,
+            )
+        ]
+
+        slabs.append(filtered)
+
+    return np.concatenate(tuple(slabs))
+
+
+def mean_stddev_distance(
+    params: _py4dgeo.DistanceUncertaintyCalculationParameters,
+) -> tuple:
+    # Calculate distance
+    distance = params.normal[0, :].dot(
+        params.workingset2.mean(axis=0) - params.workingset1.mean(axis=0)
+    )
+
+    # Calculate variances
+    variance1 = params.normal @ np.cov(params.workingset1.T) @ params.normal.T
+    variance2 = params.normal @ np.cov(params.workingset2.T) @ params.normal.T
+
+    # The structured array that describes the full uncertainty
+    uncertainty = _py4dgeo.DistanceUncertainty(
+        lodetection=1.96
+        * (
+            np.sqrt(
+                variance1 / params.workingset1.shape[0]
+                + variance2 / params.workingset2.shape[0]
+            ).item()
+            + params.registration_error
+        ),
+        spread1=np.sqrt(variance1).item(),
+        num_samples1=params.workingset1.shape[0],
+        spread2=np.sqrt(variance2).item(),
+        num_samples2=params.workingset2.shape[0],
+    )
+
+    return distance, uncertainty
+
+
+def average_pos(a, pos, div):
+    # This is an unfortunate helper, but numpy.percentile does not do
+    # the correct thing. It sometimes averages although we have an exact
+    # match for the position we are searching.
+    if len(a) % div == 0:
+        return (
+            a[int(np.floor(pos * len(a)))] + a[int(np.floor(pos * len(a))) - 1]
+        ) / 2.0
+    else:
+        return a[int(np.floor(pos * len(a)))]
+
+
+def median_iqr_distance(
+    params: _py4dgeo.DistanceUncertaintyCalculationParameters,
+) -> tuple:
+    # Calculate distributions
+    dist1 = (params.workingset1 - params.corepoint[0, :]).dot(params.normal[0, :])
+    dist2 = (params.workingset2 - params.corepoint[0, :]).dot(params.normal[0, :])
+    dist1.sort()
+    dist2.sort()
+
+    median1 = average_pos(dist1, 0.5, 2)
+    median2 = average_pos(dist2, 0.5, 2)
+    iqr1 = average_pos(dist1, 0.75, 4) - average_pos(dist1, 0.25, 4)
+    iqr2 = average_pos(dist2, 0.75, 4) - average_pos(dist2, 0.25, 4)
+
+    # The structured array that describes the full uncertainty
+    uncertainty = _py4dgeo.DistanceUncertainty(
+        lodetection=1.96
+        * (
+            np.sqrt(
+                iqr1 * iqr1 / params.workingset1.shape[0]
+                + iqr2 * iqr2 / params.workingset2.shape[0]
+            )
+            + params.registration_error
+        ),
+        spread1=iqr1,
+        num_samples1=params.workingset1.shape[0],
+        spread2=iqr2,
+        num_samples2=params.workingset2.shape[0],
+    )
+
+    return median2 - median1, uncertainty
+
+
+class PythonFallbackM3C2(M3C2):
+    """An implementation of M3C2 that makes use of Python fallback implementations"""
+
+    @property
+    def name(self):
+        return "M3C2 (Python Fallback)"
+
+    def callback_workingset_finder(self):
+        return cylinder_workingset_finder
+
+    def callback_distance_calculation(self):
+        if self.robust_aggr:
+            return median_iqr_distance
+        else:
+            return mean_stddev_distance

py4dgeo/logger.py

@@ -0,0 +1,82 @@
+import contextlib
+import logging
+import sys
+import time
+
+
+def create_default_logger(filename=None):
+    # Create the logger instance
+    logger = logging.getLogger("py4dgeo")
+
+    # Close and remove existing handlers to avoid duplication and leaks
+    for handler in logger.handlers[:]:
+        try:
+            handler.close()
+        except Exception:
+            pass
+        logger.removeHandler(handler)
+
+    # Apply default for logfile name
+    if filename is None:
+        filename = "py4dgeo.log"
+
+    # We format messages including the date
+    _formatter = logging.Formatter(
+        "[%(asctime)s][%(levelname)s] %(message)s", "%Y-%m-%d %H:%M:%S"
+    )
+
+    # We use stdout for DEBUG and INFO messages
+    _stdoutandler = logging.StreamHandler(sys.stdout)
+    _stdoutandler.setLevel(logging.DEBUG)
+    _stdoutandler.addFilter(lambda r: r.levelno <= logging.INFO)
+    _stdoutandler.setFormatter(_formatter)
+    logger.addHandler(_stdoutandler)
+
+    # We use stderr for WARNING and ERROR messages
+    _stderrhandler = logging.StreamHandler(sys.stderr)
+    _stderrhandler.setLevel(logging.WARNING)
+    _stderrhandler.setFormatter(_formatter)
+    logger.addHandler(_stderrhandler)
+
+    # We additionally use a file that is automatically generated
+    _filehandler = logging.FileHandler(filename, mode="a", delay=True)
+    _filehandler.setLevel(logging.DEBUG)
+    _filehandler.setFormatter(_formatter)
+    logger.addHandler(_filehandler)
+
+    logger.setLevel(logging.INFO)
+
+    return logger
+
+
+# Storage to keep the logger instance alive + initial creation
+_logger = create_default_logger()
+
+
+def set_py4dgeo_logfile(filename):
+    """Set the logfile used by py4dgeo
+
+    All log messages produced by py4dgeo are logged into this file
+    in addition to be logged to stdout/stderr. By default, that file
+    is called 'py4dgeo.log'.
+
+    :param filename:
+        The name of the logfile to use
+    :type filename: str
+    """
+    global _logger
+    _logger = create_default_logger(filename)
+
+
+@contextlib.contextmanager
+def logger_context(msg, level=logging.INFO):
+    # Log a message that we started the task described by message
+    logger = logging.getLogger("py4dgeo")
+    logger.log(level, f"Starting: {msg}")
+
+    # Measure time
+    start = time.perf_counter()
+    yield
+    duration = time.perf_counter() - start
+
+    logger.log(level, f"Finished in {duration:.4f}s: {msg}")

py4dgeo/m3c2.py

@@ -0,0 +1,246 @@
+from ast import List
+from py4dgeo.epoch import Epoch, as_epoch
+from py4dgeo.util import (
+    as_double_precision,
+    MemoryPolicy,
+    Py4DGeoError,
+    make_contiguous,
+    memory_policy_is_minimum,
+)
+
+import abc
+import logging
+import numpy as np
+import typing
+import laspy
+
+import _py4dgeo
+
+logger = logging.getLogger("py4dgeo")
+
+
+class M3C2LikeAlgorithm(abc.ABC):
+    def __init__(
+        self,
+        epochs: typing.Optional[typing.Tuple[Epoch, ...]] = None,
+        corepoints: typing.Optional[np.ndarray] = None,
+        cyl_radii: typing.Optional[List] = None,
+        cyl_radius: typing.Optional[float] = None,
+        max_distance: float = 0.0,
+        registration_error: float = 0.0,
+        robust_aggr: bool = False,
+    ):
+        self.epochs = epochs
+        self.corepoints = corepoints
+        self.cyl_radii = cyl_radii
+        self.cyl_radius = cyl_radius
+        self.max_distance = max_distance
+        self.registration_error = registration_error
+        self.robust_aggr = robust_aggr
+
+    @property
+    def corepoints(self):
+        return self._corepoints
+
+    @corepoints.setter
+    def corepoints(self, _corepoints):
+        if _corepoints is None:
+            self._corepoints = None
+        else:
+            if len(_corepoints.shape) != 2 or _corepoints.shape[1] != 3:
+                raise Py4DGeoError(
+                    "Corepoints need to be given as an array of shape nx3"
+                )
+            self._corepoints = as_double_precision(make_contiguous(_corepoints))
+
+    @property
+    def epochs(self):
+        return self._epochs
+
+    @epochs.setter
+    def epochs(self, _epochs):
+        if _epochs is not None and len(_epochs) != 2:
+            raise Py4DGeoError("Exactly two epochs need to be given!")
+        self._epochs = _epochs
+
+    @property
+    def name(self):
+        raise NotImplementedError
+
+    def directions(self):
+        """The normal direction(s) to use for this algorithm."""
+        raise NotImplementedError
+
+    def calculate_distances(
+        self, epoch1, epoch2, searchtree: typing.Optional[str] = None
+    ):
+        """Calculate the distances between two epochs"""
+
+        if isinstance(self.cyl_radii, typing.Iterable):
+            logger.warning(
+                "DEPRECATION: use cyl_radius instead of cyl_radii. In a future version, cyl_radii will be removed!"
+            )
+            if len(self.cyl_radii) != 1:
+                raise Py4DGeoError(
+                    "cyl_radii must be a list containing a single float!"
+                )
+            elif self.cyl_radius is None:
+                self.cyl_radius = self.cyl_radii[0]
+            self.cyl_radii = None
+
+        if self.cyl_radius is None:
+            raise Py4DGeoError(
+                f"{self.name} requires exactly one cylinder radius to be given as a float."
+            )
+
+        # Ensure appropriate trees are built
+        epoch1._validate_search_tree()
+        epoch2._validate_search_tree()
+
+        distances, uncertainties = _py4dgeo.compute_distances(
+            self.corepoints,
+            self.cyl_radius,
+            epoch1,
+            epoch2,
+            self.directions(),
+            self.max_distance,
+            self.registration_error,
+            self.callback_workingset_finder(),
+            self.callback_distance_calculation(),
+        )
+
+        return distances, uncertainties
+
+    def run(self):
+        """Main entry point for running the algorithm"""
+        return self.calculate_distances(self.epochs[0], self.epochs[1])
+
+    def callback_workingset_finder(self):
+        """The callback used to determine the point cloud subset around a corepoint"""
+        return _py4dgeo.cylinder_workingset_finder
+
+    def callback_distance_calculation(self):
+        """The callback used to calculate the distance between two point clouds"""
+        if self.robust_aggr:
+            return _py4dgeo.median_iqr_distance
+        else:
+            return _py4dgeo.mean_stddev_distance
+
+
+class M3C2(M3C2LikeAlgorithm):
+    def __init__(
+        self,
+        normal_radii: typing.List[float] = None,
+        orientation_vector: np.ndarray = np.array([0, 0, 1]),
+        corepoint_normals: np.ndarray = None,
+        cloud_for_normals: Epoch = None,
+        **kwargs,
+    ):
+        self.normal_radii = normal_radii
+        self.orientation_vector = as_double_precision(
+            make_contiguous(orientation_vector), policy_check=False
+        )
+        self.cloud_for_normals = cloud_for_normals
+        self.corepoint_normals = corepoint_normals
+        self._directions_radii = None
+        super().__init__(**kwargs)
+
+    def directions(self):
+        # If we already have normals, we return them. This happens e.g. if the user
+        # explicitly provided them or if we already computed them in a previous run.
+        if self.corepoint_normals is not None:
+            # Make sure that the normals use double precision
+            self.corepoint_normals = as_double_precision(self.corepoint_normals)
+
+            # Assert that the normal array has the correct shape
+            if (
+                len(self.corepoint_normals.shape) != 2
+                or self.corepoint_normals.shape[0] not in (1, self.corepoints.shape[0])
+                or self.corepoint_normals.shape[1] != 3
+            ):
+                raise Py4DGeoError(
+                    f"Incompative size of corepoint normal array {self.corepoint_normals.shape}, expected {self.corepoints.shape} or (1, 3)!"
+                )
+
+            return self.corepoint_normals
+
+        # This does not work in STRICT mode
+        if not memory_policy_is_minimum(MemoryPolicy.MINIMAL):
+            raise Py4DGeoError(
+                "M3C2 requires at least the MINIMUM memory policy level to compute multiscale normals"
+            )
+
+        # Find the correct epoch to use for normal calculation
+        normals_epoch = self.cloud_for_normals
+        if normals_epoch is None:
+            normals_epoch = self.epochs[0]
+        normals_epoch = as_epoch(normals_epoch)
+
+        # Ensure appropriate tree structures have been built
+        normals_epoch._validate_search_tree()
+
+        # Trigger the precomputation
+        self.corepoint_normals, self._directions_radii = (
+            _py4dgeo.compute_multiscale_directions(
+                normals_epoch,
+                self.corepoints,
+                self.normal_radii,
+                self.orientation_vector,
+            )
+        )
+
+        return self.corepoint_normals
+
+    def directions_radii(self):
+        if self._directions_radii is None:
+            raise ValueError(
+                "Radii are only available after calculating directions with py4dgeo."
+            )
+
+        return self._directions_radii
+
+    @property
+    def name(self):
+        return "M3C2"
+
+
+def write_m3c2_results_to_las(
+    outfilepath: str, m3c2: M3C2LikeAlgorithm, attribute_dict: dict = {}
+):
+    """Save the corepoints, distances and other attributes to a given las filename
+
+    :param outfilepath:
+        The las file path to save the corepoints, distances and other attributes.
+    :type outfilepath: str
+    :param m3c2:
+        The M3C2LikeAlgorithm object.
+    :type m3c2: M3C2LikeAlgorithm
+    :param attribute_dict:
+        The dictionary of attributes which will be saved together with corepoints.
+    :type attribute_dict: dict
+    """
+    # Will utilize Epoch.save(), by creating epoch from m3c2 corepoints and attribute_dict
+    # to be the epoch additional_dimensions, write this epoch to las not a zip file.
+    outpoints = m3c2.corepoints
+    hdr = laspy.LasHeader(version="1.4", point_format=6)
+    hdr.x_scale = 0.00025
+    hdr.y_scale = 0.00025
+    hdr.z_scale = 0.00025
+    mean_extent = np.mean(outpoints, axis=0)
+    hdr.x_offset = int(mean_extent[0])
+    hdr.y_offset = int(mean_extent[1])
+    hdr.z_offset = int(mean_extent[2])
+
+    las = laspy.LasData(hdr)
+
+    las.x = outpoints[:, 0]
+    las.y = outpoints[:, 1]
+    las.z = outpoints[:, 2]
+    for key, vals in attribute_dict.items():
+        try:
+            las[key] = vals
+        except:
+            las.add_extra_dim(laspy.ExtraBytesParams(name=key, type=type(vals[0])))
+            las[key] = vals
+
+    las.write(outfilepath)