Upload 879 files

.gitattributes

@@ -35,3 +35,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 color_features.json filter=lfs diff=lfs merge=lfs -text
 image_subset.json filter=lfs diff=lfs merge=lfs -text
+skimage/filters/rank/generic_cy.cp310-win_amd64.pyd filter=lfs diff=lfs merge=lfs -text

skimage/__init__.py

@@ -0,0 +1,182 @@
+"""Image Processing for Python
+
+``scikit-image`` (a.k.a. ``skimage``) is a collection of algorithms for image
+processing and computer vision.
+
+The main package of ``skimage`` only provides a few utilities for converting
+between image data types; for most features, you need to import one of the
+following subpackages:
+
+Subpackages
+-----------
+color
+    Color space conversion.
+data
+    Test images and example data.
+draw
+    Drawing primitives (lines, text, etc.) that operate on NumPy arrays.
+exposure
+    Image intensity adjustment, e.g., histogram equalization, etc.
+feature
+    Feature detection and extraction, e.g., texture analysis corners, etc.
+filters
+    Sharpening, edge finding, rank filters, thresholding, etc.
+graph
+    Graph-theoretic operations, e.g., shortest paths.
+io
+    Reading, saving, and displaying images and video.
+measure
+    Measurement of image properties, e.g., region properties and contours.
+metrics
+    Metrics corresponding to images, e.g. distance metrics, similarity, etc.
+morphology
+    Morphological operations, e.g., opening or skeletonization.
+restoration
+    Restoration algorithms, e.g., deconvolution algorithms, denoising, etc.
+segmentation
+    Partitioning an image into multiple regions.
+transform
+    Geometric and other transforms, e.g., rotation or the Radon transform.
+util
+    Generic utilities.
+
+Utility Functions
+-----------------
+img_as_float
+    Convert an image to floating point format, with values in [0, 1].
+    Is similar to `img_as_float64`, but will not convert lower-precision
+    floating point arrays to `float64`.
+img_as_float32
+    Convert an image to single-precision (32-bit) floating point format,
+    with values in [0, 1].
+img_as_float64
+    Convert an image to double-precision (64-bit) floating point format,
+    with values in [0, 1].
+img_as_uint
+    Convert an image to unsigned integer format, with values in [0, 65535].
+img_as_int
+    Convert an image to signed integer format, with values in [-32768, 32767].
+img_as_ubyte
+    Convert an image to unsigned byte format, with values in [0, 255].
+img_as_bool
+    Convert an image to boolean format, with values either True or False.
+dtype_limits
+    Return intensity limits, i.e. (min, max) tuple, of the image's dtype.
+
+"""
+
+__version__ = '0.23.2'
+
+import lazy_loader as lazy
+
+__getattr__, __lazy_dir__, _ = lazy.attach_stub(__name__, __file__)
+
+
+def __dir__():
+    return __lazy_dir__() + ['__version__']
+
+
+# Logic for checking for improper install and importing while in the source
+# tree when package has not been installed inplace.
+# Code adapted from scikit-learn's __check_build module.
+_INPLACE_MSG = """
+It appears that you are importing a local scikit-image source tree. For
+this, you need to have an inplace install. Maybe you are in the source
+directory and you need to try from another location."""
+
+_STANDARD_MSG = """
+Your install of scikit-image appears to be broken.
+Try re-installing the package following the instructions at:
+https://scikit-image.org/docs/stable/user_guide/install.html"""
+
+
+def _raise_build_error(e):
+    # Raise a comprehensible error
+    import os.path as osp
+
+    local_dir = osp.split(__file__)[0]
+    msg = _STANDARD_MSG
+    if local_dir == "skimage":
+        # Picking up the local install: this will work only if the
+        # install is an 'inplace build'
+        msg = _INPLACE_MSG
+    raise ImportError(
+        f"{e}\nIt seems that scikit-image has not been built correctly.\n{msg}"
+    )
+
+
+try:
+    # This variable is injected in the __builtins__ by the build
+    # process. It used to enable importing subpackages of skimage when
+    # the binaries are not built
+    __SKIMAGE_SETUP__
+except NameError:
+    __SKIMAGE_SETUP__ = False
+
+if __SKIMAGE_SETUP__:
+    import sys
+
+    sys.stderr.write('Partial import of skimage during the build process.\n')
+    # We are not importing the rest of the scikit during the build
+    # process, as it may not be compiled yet
+else:
+    try:
+        from ._shared import geometry
+
+        del geometry
+    except ImportError as e:
+        _raise_build_error(e)
+
+    # Legacy imports into the root namespace; not advertised in __all__
+    from .util.dtype import (
+        dtype_limits,
+        img_as_float32,
+        img_as_float64,
+        img_as_float,
+        img_as_int,
+        img_as_uint,
+        img_as_ubyte,
+        img_as_bool,
+    )
+
+    from .util.lookfor import lookfor
+
+    from .data import data_dir
+
+
+if 'dev' in __version__:
+    # Append last commit date and hash to dev version information, if available
+
+    import subprocess
+    import os.path
+
+    try:
+        p = subprocess.Popen(
+            ['git', 'log', '-1', '--format="%h %aI"'],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            cwd=os.path.dirname(__file__),
+        )
+    except FileNotFoundError:
+        pass
+    else:
+        out, err = p.communicate()
+        if p.returncode == 0:
+            git_hash, git_date = (
+                out.decode('utf-8')
+                .strip()
+                .replace('"', '')
+                .split('T')[0]
+                .replace('-', '')
+                .split()
+            )
+
+            __version__ = '+'.join(
+                [tag for tag in __version__.split('+') if not tag.startswith('git')]
+            )
+            __version__ += f'+git{git_date}.{git_hash}'
+
+from skimage._shared.tester import PytestTester  # noqa
+
+test = PytestTester(__name__)
+del PytestTester

skimage/__init__.pyi

@@ -0,0 +1,41 @@
+submodules = [
+    'color',
+    'data',
+    'draw',
+    'exposure',
+    'feature',
+    'filters',
+    'future',
+    'graph',
+    'io',
+    'measure',
+    'metrics',
+    'morphology',
+    'registration',
+    'restoration',
+    'segmentation',
+    'transform',
+    'util',
+]
+
+__all__ = submodules + ['__version__']  # noqa: F822
+
+from . import (
+    color,
+    data,
+    draw,
+    exposure,
+    feature,
+    filters,
+    future,
+    graph,
+    io,
+    measure,
+    metrics,
+    morphology,
+    registration,
+    restoration,
+    segmentation,
+    transform,
+    util,
+)

skimage/_shared/_dependency_checks.py

@@ -0,0 +1,3 @@
+from .version_requirements import is_installed
+
+has_mpl = is_installed("matplotlib", ">=3.3")

skimage/_shared/_geometry.py

@@ -0,0 +1,54 @@
+__all__ = ['polygon_clip', 'polygon_area']
+
+import numpy as np
+
+from .version_requirements import require
+
+
+@require("matplotlib", ">=3.3")
+def polygon_clip(rp, cp, r0, c0, r1, c1):
+    """Clip a polygon to the given bounding box.
+
+    Parameters
+    ----------
+    rp, cp : (K,) ndarray of double
+        Row and column coordinates of the polygon.
+    (r0, c0), (r1, c1) : double
+        Top-left and bottom-right coordinates of the bounding box.
+
+    Returns
+    -------
+    r_clipped, c_clipped : (L,) ndarray of double
+        Coordinates of clipped polygon.
+
+    Notes
+    -----
+    This makes use of Sutherland-Hodgman clipping as implemented in
+    AGG 2.4 and exposed in Matplotlib.
+
+    """
+    from matplotlib import path, transforms
+
+    poly = path.Path(np.vstack((rp, cp)).T, closed=True)
+    clip_rect = transforms.Bbox([[r0, c0], [r1, c1]])
+    poly_clipped = poly.clip_to_bbox(clip_rect).to_polygons()[0]
+
+    return poly_clipped[:, 0], poly_clipped[:, 1]
+
+
+def polygon_area(pr, pc):
+    """Compute the area of a polygon.
+
+    Parameters
+    ----------
+    pr, pc : (K,) array of float
+        Polygon row and column coordinates.
+
+    Returns
+    -------
+    a : float
+        Area of the polygon.
+    """
+    pr = np.asarray(pr)
+    pc = np.asarray(pc)
+    return 0.5 * np.abs(np.sum((pc[:-1] * pr[1:]) - (pc[1:] * pr[:-1])))

skimage/_shared/_tempfile.py

@@ -0,0 +1,28 @@
+from tempfile import NamedTemporaryFile
+from contextlib import contextmanager
+import os
+
+
+@contextmanager
+def temporary_file(suffix=''):
+    """Yield a writeable temporary filename that is deleted on context exit.
+
+    Parameters
+    ----------
+    suffix : string, optional
+        The suffix for the file.
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from skimage import io
+    >>> with temporary_file('.tif') as tempfile:
+    ...     im = np.arange(25, dtype=np.uint8).reshape((5, 5))
+    ...     io.imsave(tempfile, im)
+    ...     assert np.all(io.imread(tempfile) == im)
+    """
+    with NamedTemporaryFile(suffix=suffix, delete=False) as tempfile_stream:
+        tempfile = tempfile_stream.name
+
+    yield tempfile
+    os.remove(tempfile)

skimage/_shared/_warnings.py

@@ -0,0 +1,149 @@
+from contextlib import contextmanager
+import sys
+import warnings
+import re
+import functools
+import os
+
+__all__ = ['all_warnings', 'expected_warnings', 'warn']
+
+
+# A version of `warnings.warn` with a default stacklevel of 2.
+# functool is used so as not to increase the call stack accidentally
+warn = functools.partial(warnings.warn, stacklevel=2)
+
+
+@contextmanager
+def all_warnings():
+    """
+    Context for use in testing to ensure that all warnings are raised.
+
+    Examples
+    --------
+    >>> import warnings
+    >>> def foo():
+    ...     warnings.warn(RuntimeWarning("bar"), stacklevel=2)
+
+    We raise the warning once, while the warning filter is set to "once".
+    Hereafter, the warning is invisible, even with custom filters:
+
+    >>> with warnings.catch_warnings():
+    ...     warnings.simplefilter('once')
+    ...     foo()                         # doctest: +SKIP
+
+    We can now run ``foo()`` without a warning being raised:
+
+    >>> from numpy.testing import assert_warns
+    >>> foo()                             # doctest: +SKIP
+
+    To catch the warning, we call in the help of ``all_warnings``:
+
+    >>> with all_warnings():
+    ...     assert_warns(RuntimeWarning, foo)
+    """
+    # _warnings.py is on the critical import path.
+    # Since this is a testing only function, we lazy import inspect.
+    import inspect
+
+    # Whenever a warning is triggered, Python adds a __warningregistry__
+    # member to the *calling* module.  The exercise here is to find
+    # and eradicate all those breadcrumbs that were left lying around.
+    #
+    # We proceed by first searching all parent calling frames and explicitly
+    # clearing their warning registries (necessary for the doctests above to
+    # pass).  Then, we search for all submodules of skimage and clear theirs
+    # as well (necessary for the skimage test suite to pass).
+
+    frame = inspect.currentframe()
+    if frame:
+        for f in inspect.getouterframes(frame):
+            f[0].f_locals['__warningregistry__'] = {}
+    del frame
+
+    for mod_name, mod in list(sys.modules.items()):
+        try:
+            mod.__warningregistry__.clear()
+        except AttributeError:
+            pass
+
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        yield w
+
+
+@contextmanager
+def expected_warnings(matching):
+    r"""Context for use in testing to catch known warnings matching regexes
+
+    Parameters
+    ----------
+    matching : None or a list of strings or compiled regexes
+        Regexes for the desired warning to catch
+        If matching is None, this behaves as a no-op.
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> rng = np.random.default_rng()
+    >>> image = rng.integers(0, 2**16, size=(100, 100), dtype=np.uint16)
+    >>> # rank filters are slow when bit-depth exceeds 10 bits
+    >>> from skimage import filters
+    >>> with expected_warnings(['Bad rank filter performance']):
+    ...     median_filtered = filters.rank.median(image)
+
+    Notes
+    -----
+    Uses `all_warnings` to ensure all warnings are raised.
+    Upon exiting, it checks the recorded warnings for the desired matching
+    pattern(s).
+    Raises a ValueError if any match was not found or an unexpected
+    warning was raised.
+    Allows for three types of behaviors: `and`, `or`, and `optional` matches.
+    This is done to accommodate different build environments or loop conditions
+    that may produce different warnings.  The behaviors can be combined.
+    If you pass multiple patterns, you get an orderless `and`, where all of the
+    warnings must be raised.
+    If you use the `|` operator in a pattern, you can catch one of several
+    warnings.
+    Finally, you can use `|\A\Z` in a pattern to signify it as optional.
+
+    """
+    if isinstance(matching, str):
+        raise ValueError(
+            '``matching`` should be a list of strings and not ' 'a string itself.'
+        )
+
+    # Special case for disabling the context manager
+    if matching is None:
+        yield None
+        return
+
+    strict_warnings = os.environ.get('SKIMAGE_TEST_STRICT_WARNINGS', '1')
+    if strict_warnings.lower() == 'true':
+        strict_warnings = True
+    elif strict_warnings.lower() == 'false':
+        strict_warnings = False
+    else:
+        strict_warnings = bool(int(strict_warnings))
+
+    with all_warnings() as w:
+        # enter context
+        yield w
+        # exited user context, check the recorded warnings
+        # Allow users to provide None
+        while None in matching:
+            matching.remove(None)
+        remaining = [m for m in matching if r'\A\Z' not in m.split('|')]
+        for warn in w:
+            found = False
+            for match in matching:
+                if re.search(match, str(warn.message)) is not None:
+                    found = True
+                    if match in remaining:
+                        remaining.remove(match)
+            if strict_warnings and not found:
+                raise ValueError(f'Unexpected warning: {str(warn.message)}')
+        if strict_warnings and (len(remaining) > 0):
+            newline = "\n"
+            msg = f"No warning raised matching:{newline}{newline.join(remaining)}"
+            raise ValueError(msg)

skimage/_shared/compat.py

@@ -0,0 +1,30 @@
+"""Compatibility helpers for dependencies."""
+
+from packaging.version import parse
+
+import numpy as np
+import scipy as sp
+
+
+__all__ = [
+    "NP_COPY_IF_NEEDED",
+    "SCIPY_CG_TOL_PARAM_NAME",
+]
+
+
+NUMPY_LT_2_0_0 = parse(np.__version__) < parse('2.0.0.dev0')
+
+# With NumPy 2.0.0, `copy=False` now raises a ValueError if the copy cannot be
+# made. The previous behavior to only copy if needed is provided with `copy=None`.
+# During the transition period, use this symbol instead.
+# Remove once NumPy 2.0.0 is the minimal required version.
+# https://numpy.org/devdocs/release/2.0.0-notes.html#new-copy-keyword-meaning-for-array-and-asarray-constructors
+# https://github.com/numpy/numpy/pull/25168
+NP_COPY_IF_NEEDED = False if NUMPY_LT_2_0_0 else None
+
+
+SCIPY_LT_1_12 = parse(sp.__version__) < parse('1.12')
+
+# Starting in SciPy v1.12, 'scipy.sparse.linalg.cg' keyword argument `tol` is
+# deprecated in favor of `rtol`.
+SCIPY_CG_TOL_PARAM_NAME = "tol" if SCIPY_LT_1_12 else "rtol"

skimage/_shared/coord.py

@@ -0,0 +1,125 @@
+import numpy as np
+from scipy.spatial import cKDTree, distance
+
+
+def _ensure_spacing(coord, spacing, p_norm, max_out):
+    """Returns a subset of coord where a minimum spacing is guaranteed.
+
+    Parameters
+    ----------
+    coord : ndarray
+        The coordinates of the considered points.
+    spacing : float
+        the maximum allowed spacing between the points.
+    p_norm : float
+        Which Minkowski p-norm to use. Should be in the range [1, inf].
+        A finite large p may cause a ValueError if overflow can occur.
+        ``inf`` corresponds to the Chebyshev distance and 2 to the
+        Euclidean distance.
+    max_out: int
+        If not None, at most the first ``max_out`` candidates are
+        returned.
+
+    Returns
+    -------
+    output : ndarray
+        A subset of coord where a minimum spacing is guaranteed.
+
+    """
+
+    # Use KDtree to find the peaks that are too close to each other
+    tree = cKDTree(coord)
+
+    indices = tree.query_ball_point(coord, r=spacing, p=p_norm)
+    rejected_peaks_indices = set()
+    naccepted = 0
+    for idx, candidates in enumerate(indices):
+        if idx not in rejected_peaks_indices:
+            # keep current point and the points at exactly spacing from it
+            candidates.remove(idx)
+            dist = distance.cdist(
+                [coord[idx]], coord[candidates], distance.minkowski, p=p_norm
+            ).reshape(-1)
+            candidates = [c for c, d in zip(candidates, dist) if d < spacing]
+
+            # candidates.remove(keep)
+            rejected_peaks_indices.update(candidates)
+            naccepted += 1
+            if max_out is not None and naccepted >= max_out:
+                break
+
+    # Remove the peaks that are too close to each other
+    output = np.delete(coord, tuple(rejected_peaks_indices), axis=0)
+    if max_out is not None:
+        output = output[:max_out]
+
+    return output
+
+
+def ensure_spacing(
+    coords,
+    spacing=1,
+    p_norm=np.inf,
+    min_split_size=50,
+    max_out=None,
+    *,
+    max_split_size=2000,
+):
+    """Returns a subset of coord where a minimum spacing is guaranteed.
+
+    Parameters
+    ----------
+    coords : array_like
+        The coordinates of the considered points.
+    spacing : float
+        the maximum allowed spacing between the points.
+    p_norm : float
+        Which Minkowski p-norm to use. Should be in the range [1, inf].
+        A finite large p may cause a ValueError if overflow can occur.
+        ``inf`` corresponds to the Chebyshev distance and 2 to the
+        Euclidean distance.
+    min_split_size : int
+        Minimum split size used to process ``coords`` by batch to save
+        memory. If None, the memory saving strategy is not applied.
+    max_out : int
+        If not None, only the first ``max_out`` candidates are returned.
+    max_split_size : int
+        Maximum split size used to process ``coords`` by batch to save
+        memory. This number was decided by profiling with a large number
+        of points. Too small a number results in too much looping in
+        Python instead of C, slowing down the process, while too large
+        a number results in large memory allocations, slowdowns, and,
+        potentially, in the process being killed -- see gh-6010. See
+        benchmark results `here
+        <https://github.com/scikit-image/scikit-image/pull/6035#discussion_r751518691>`_.
+
+    Returns
+    -------
+    output : array_like
+        A subset of coord where a minimum spacing is guaranteed.
+
+    """
+
+    output = coords
+    if len(coords):
+        coords = np.atleast_2d(coords)
+        if min_split_size is None:
+            batch_list = [coords]
+        else:
+            coord_count = len(coords)
+            split_idx = [min_split_size]
+            split_size = min_split_size
+            while coord_count - split_idx[-1] > max_split_size:
+                split_size *= 2
+                split_idx.append(split_idx[-1] + min(split_size, max_split_size))
+            batch_list = np.array_split(coords, split_idx)
+
+        output = np.zeros((0, coords.shape[1]), dtype=coords.dtype)
+        for batch in batch_list:
+            output = _ensure_spacing(
+                np.vstack([output, batch]), spacing, p_norm, max_out
+            )
+            if max_out is not None and len(output) >= max_out:
+                break
+
+    return output

skimage/_shared/dtype.py

@@ -0,0 +1,73 @@
+import numpy as np
+
+# Define classes of supported dtypes and Python scalar types
+# Variables ending in `_dtypes` only contain numpy.dtypes of the respective
+# class; variables ending in `_types` additionally include Python scalar types.
+signed_integer_dtypes = {np.int8, np.int16, np.int32, np.int64}
+signed_integer_types = signed_integer_dtypes | {int}
+
+unsigned_integer_dtypes = {np.uint8, np.uint16, np.uint32, np.uint64}
+
+integer_dtypes = signed_integer_dtypes | unsigned_integer_dtypes
+integer_types = signed_integer_types | unsigned_integer_dtypes
+
+floating_dtypes = {np.float16, np.float32, np.float64}
+floating_types = floating_dtypes | {float}
+
+complex_dtypes = {np.complex64, np.complex128}
+complex_types = complex_dtypes | {complex}
+
+inexact_dtypes = floating_dtypes | complex_dtypes
+inexact_types = floating_types | complex_types
+
+bool_types = {np.dtype(bool), bool}
+
+numeric_dtypes = integer_dtypes | inexact_dtypes | {np.bool_}
+numeric_types = integer_types | inexact_types | bool_types
+
+
+def numeric_dtype_min_max(dtype):
+    """Return minimum and maximum representable value for a given dtype.
+
+    A convenient wrapper around `numpy.finfo` and `numpy.iinfo` that
+    additionally supports numpy.bool as well.
+
+    Parameters
+    ----------
+    dtype : numpy.dtype
+        The dtype. Tries to convert Python "types" such as int or float, to
+        the corresponding NumPy dtype.
+
+    Returns
+    -------
+    min, max : number
+        Minimum and maximum of the given `dtype`. These scalars are themselves
+        of the given `dtype`.
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> numeric_dtype_min_max(np.uint8)
+    (0, 255)
+    >>> numeric_dtype_min_max(bool)
+    (False, True)
+    >>> numeric_dtype_min_max(np.float64)
+    (-1.7976931348623157e+308, 1.7976931348623157e+308)
+    >>> numeric_dtype_min_max(int)
+    (-9223372036854775808, 9223372036854775807)
+    """
+    dtype = np.dtype(dtype)
+    if np.issubdtype(dtype, np.integer):
+        info = np.iinfo(dtype)
+        min_ = dtype.type(info.min)
+        max_ = dtype.type(info.max)
+    elif np.issubdtype(dtype, np.inexact):
+        info = np.finfo(dtype)
+        min_ = info.min
+        max_ = info.max
+    elif np.issubdtype(dtype, np.dtype(bool)):
+        min_ = dtype.type(False)
+        max_ = dtype.type(True)
+    else:
+        raise ValueError(f"unsupported dtype {dtype!r}")
+    return min_, max_

skimage/_shared/fast_exp.h

@@ -0,0 +1,47 @@
+/* A fast approximation of the exponential function.
+ * Reference [1]: https://schraudolph.org/pubs/Schraudolph99.pdf
+ * Reference [2]: https://doi.org/10.1162/089976600300015033
+ * Additional improvements by Leonid Bloch. */
+
+#include <stdint.h>
+
+/* use just EXP_A = 1512775 for integer version, to avoid FP calculations */
+#define EXP_A (1512775.3951951856938)  /* 2^20/ln2 */
+/* For min. RMS error */
+#define EXP_BC 1072632447              /* 1023*2^20 - 60801 */
+/* For min. max. relative error */
+/* #define EXP_BC 1072647449 */        /* 1023*2^20 - 45799 */
+/* For min. mean relative error */
+/* #define EXP_BC 1072625005 */        /* 1023*2^20 - 68243 */
+
+__inline double _fast_exp (double y)
+{
+    union
+    {
+        double d;
+        struct { int32_t i, j; } n;
+        char t[8];
+    } _eco;
+
+    _eco.n.i = 1;
+
+    switch(_eco.t[0]) {
+        case 1:
+            /* Little endian */
+            _eco.n.j = (int32_t)(EXP_A*(y)) + EXP_BC;
+            _eco.n.i = 0;
+            break;
+        case 0:
+            /* Big endian */
+            _eco.n.i = (int32_t)(EXP_A*(y)) + EXP_BC;
+            _eco.n.j = 0;
+            break;
+    }
+
+    return _eco.d;
+}
+
+__inline float _fast_expf (float y)
+{
+  return (float)_fast_exp((double)y);
+}

skimage/_shared/filters.py

@@ -0,0 +1,142 @@
+"""Filters used across multiple skimage submodules.
+
+These are defined here to avoid circular imports.
+
+The unit tests remain under skimage/filters/tests/
+"""
+
+from collections.abc import Iterable
+
+import numpy as np
+from scipy import ndimage as ndi
+
+from .._shared.utils import (
+    _supported_float_type,
+    convert_to_float,
+    deprecate_parameter,
+    DEPRECATED,
+)
+
+
+@deprecate_parameter(
+    "output", new_name="out", start_version="0.23", stop_version="0.25"
+)
+def gaussian(
+    image,
+    sigma=1,
+    output=DEPRECATED,
+    mode='nearest',
+    cval=0,
+    preserve_range=False,
+    truncate=4.0,
+    *,
+    channel_axis=None,
+    out=None,
+):
+    """Multi-dimensional Gaussian filter.
+
+    Parameters
+    ----------
+    image : ndarray
+        Input image (grayscale or color) to filter.
+    sigma : scalar or sequence of scalars, optional
+        Standard deviation for Gaussian kernel. The standard
+        deviations of the Gaussian filter are given for each axis as a
+        sequence, or as a single number, in which case it is equal for
+        all axes.
+    mode : {'reflect', 'constant', 'nearest', 'mirror', 'wrap'}, optional
+        The ``mode`` parameter determines how the array borders are
+        handled, where ``cval`` is the value when mode is equal to
+        'constant'. Default is 'nearest'.
+    cval : scalar, optional
+        Value to fill past edges of input if ``mode`` is 'constant'. Default
+        is 0.0
+    preserve_range : bool, optional
+        If True, keep the original range of values. Otherwise, the input
+        ``image`` is converted according to the conventions of ``img_as_float``
+        (Normalized first to values [-1.0 ; 1.0] or [0 ; 1.0] depending on
+        dtype of input)
+
+        For more information, see:
+        https://scikit-image.org/docs/dev/user_guide/data_types.html
+    truncate : float, optional
+        Truncate the filter at this many standard deviations.
+    channel_axis : int or None, optional
+        If None, the image is assumed to be a grayscale (single channel) image.
+        Otherwise, this parameter indicates which axis of the array corresponds
+        to channels.
+
+        .. versionadded:: 0.19
+           `channel_axis` was added in 0.19.
+    out : ndarray, optional
+        If given, the filtered image will be stored in this array.
+
+        .. versionadded:: 0.23
+            `out` was added in 0.23.
+
+    Returns
+    -------
+    filtered_image : ndarray
+        the filtered array
+
+    Notes
+    -----
+    This function is a wrapper around :func:`scipy.ndimage.gaussian_filter`.
+
+    Integer arrays are converted to float.
+
+    `out` should be of floating-point data type since `gaussian` converts the
+    input `image` to float. If `out` is not provided, another array
+    will be allocated and returned as the result.
+
+    The multi-dimensional filter is implemented as a sequence of
+    one-dimensional convolution filters. The intermediate arrays are
+    stored in the same data type as the output. Therefore, for output
+    types with a limited precision, the results may be imprecise
+    because intermediate results may be stored with insufficient
+    precision.
+
+    Examples
+    --------
+    >>> import skimage as ski
+    >>> a = np.zeros((3, 3))
+    >>> a[1, 1] = 1
+    >>> a
+    array([[0., 0., 0.],
+           [0., 1., 0.],
+           [0., 0., 0.]])
+    >>> ski.filters.gaussian(a, sigma=0.4)  # mild smoothing
+    array([[0.00163116, 0.03712502, 0.00163116],
+           [0.03712502, 0.84496158, 0.03712502],
+           [0.00163116, 0.03712502, 0.00163116]])
+    >>> ski.filters.gaussian(a, sigma=1)  # more smoothing
+    array([[0.05855018, 0.09653293, 0.05855018],
+           [0.09653293, 0.15915589, 0.09653293],
+           [0.05855018, 0.09653293, 0.05855018]])
+    >>> # Several modes are possible for handling boundaries
+    >>> ski.filters.gaussian(a, sigma=1, mode='reflect')
+    array([[0.08767308, 0.12075024, 0.08767308],
+           [0.12075024, 0.16630671, 0.12075024],
+           [0.08767308, 0.12075024, 0.08767308]])
+    >>> # For RGB images, each is filtered separately
+    >>> image = ski.data.astronaut()
+    >>> filtered_img = ski.filters.gaussian(image, sigma=1, channel_axis=-1)
+
+    """
+    if np.any(np.asarray(sigma) < 0.0):
+        raise ValueError("Sigma values less than zero are not valid")
+    if channel_axis is not None:
+        # do not filter across channels
+        if not isinstance(sigma, Iterable):
+            sigma = [sigma] * (image.ndim - 1)
+        if len(sigma) == image.ndim - 1:
+            sigma = list(sigma)
+            sigma.insert(channel_axis % image.ndim, 0)
+    image = convert_to_float(image, preserve_range)
+    float_dtype = _supported_float_type(image.dtype)
+    image = image.astype(float_dtype, copy=False)
+    if (out is not None) and (not np.issubdtype(out.dtype, np.floating)):
+        raise ValueError(f"dtype of `out` must be float; got {out.dtype!r}.")
+    return ndi.gaussian_filter(
+        image, sigma, output=out, mode=mode, cval=cval, truncate=truncate
+    )

skimage/_shared/tester.py

@@ -0,0 +1,131 @@
+import os
+import sys
+
+
+def _show_skimage_info():
+    import skimage
+
+    print(f"skimage version {skimage.__version__}")
+
+
+class PytestTester:
+    """
+    Pytest test runner.
+
+    This class is made available in ``skimage._shared.testing``, and a test
+    function is typically added to a package's __init__.py like so::
+
+      from skimage._shared.testing import PytestTester
+      test = PytestTester(__name__)
+      del PytestTester
+
+    Calling this test function finds and runs all tests associated with the
+    module and all its sub-modules.
+
+    Attributes
+    ----------
+    module_name : str
+        Full path to the package to test.
+
+    Parameters
+    ----------
+    module_name : module name
+        The name of the module to test.
+
+    """
+
+    def __init__(self, module_name):
+        self.module_name = module_name
+
+    def __call__(
+        self,
+        label='fast',
+        verbose=1,
+        extra_argv=None,
+        doctests=False,
+        coverage=False,
+        durations=-1,
+        tests=None,
+    ):
+        """
+        Run tests for module using pytest.
+
+        Parameters
+        ----------
+        label : {'fast', 'full'}, optional
+            Identifies the tests to run. When set to 'fast', tests decorated
+            with `pytest.mark.slow` are skipped, when 'full', the slow marker
+            is ignored.
+        verbose : int, optional
+            Verbosity value for test outputs, in the range 1-3. Default is 1.
+        extra_argv : list, optional
+            List with any extra arguments to pass to pytests.
+        doctests : bool, optional
+            .. note:: Not supported
+        coverage : bool, optional
+            If True, report coverage of scikit-image code. Default is False.
+            Requires installation of (pip) pytest-cov.
+        durations : int, optional
+            If < 0, do nothing, If 0, report time of all tests, if > 0,
+            report the time of the slowest `timer` tests. Default is -1.
+        tests : test or list of tests
+            Tests to be executed with pytest '--pyargs'
+
+        Returns
+        -------
+        result : bool
+            Return True on success, false otherwise.
+        """
+        import pytest
+
+        module = sys.modules[self.module_name]
+        module_path = os.path.abspath(module.__path__[0])
+
+        # setup the pytest arguments
+        pytest_args = ["-l"]
+
+        # offset verbosity. The "-q" cancels a "-v".
+        pytest_args += ["-q"]
+
+        # Filter out annoying import messages. Want these in both develop and
+        # release mode.
+        pytest_args += [
+            "-W ignore:Not importing directory",
+            "-W ignore:numpy.dtype size changed",
+            "-W ignore:numpy.ufunc size changed",
+        ]
+
+        if doctests:
+            raise ValueError("Doctests not supported")
+
+        if extra_argv:
+            pytest_args += list(extra_argv)
+
+        if verbose > 1:
+            pytest_args += ["-" + "v" * (verbose - 1)]
+
+        if coverage:
+            pytest_args += ["--cov=" + module_path]
+
+        if label == "fast":
+            pytest_args += ["-m", "not slow"]
+        elif label != "full":
+            pytest_args += ["-m", label]
+
+        if durations >= 0:
+            pytest_args += [f"--durations={durations}"]
+
+        if tests is None:
+            tests = [self.module_name]
+
+        pytest_args += ["--pyargs"] + list(tests)
+
+        # run tests.
+        _show_skimage_info()
+
+        try:
+            code = pytest.main(pytest_args)
+        except SystemExit as exc:
+            code = exc.code
+
+        return code == 0

skimage/_shared/testing.py

@@ -0,0 +1,291 @@
+"""
+Testing utilities.
+"""
+
+import os
+import re
+import struct
+import threading
+import functools
+import inspect
+from tempfile import NamedTemporaryFile
+
+import numpy as np
+from numpy import testing
+from numpy.testing import (
+    TestCase,
+    assert_,
+    assert_warns,
+    assert_no_warnings,
+    assert_equal,
+    assert_almost_equal,
+    assert_array_equal,
+    assert_allclose,
+    assert_array_almost_equal,
+    assert_array_almost_equal_nulp,
+    assert_array_less,
+)
+
+from .. import data, io
+from ..data._fetchers import _fetch
+from ..util import img_as_uint, img_as_float, img_as_int, img_as_ubyte
+from ._warnings import expected_warnings
+
+import pytest
+
+
+skipif = pytest.mark.skipif
+xfail = pytest.mark.xfail
+parametrize = pytest.mark.parametrize
+raises = pytest.raises
+fixture = pytest.fixture
+
+SKIP_RE = re.compile(r"(\s*>>>.*?)(\s*)#\s*skip\s+if\s+(.*)$")
+
+
+# true if python is running in 32bit mode
+# Calculate the size of a void * pointer in bits
+# https://docs.python.org/3/library/struct.html
+arch32 = struct.calcsize("P") * 8 == 32
+
+
+def assert_less(a, b, msg=None):
+    message = f"{a!r} is not lower than {b!r}"
+    if msg is not None:
+        message += ": " + msg
+    assert a < b, message
+
+
+def assert_greater(a, b, msg=None):
+    message = f"{a!r} is not greater than {b!r}"
+    if msg is not None:
+        message += ": " + msg
+    assert a > b, message
+
+
+def doctest_skip_parser(func):
+    """Decorator replaces custom skip test markup in doctests
+
+    Say a function has a docstring::
+
+        >>> something, HAVE_AMODULE, HAVE_BMODULE = 0, False, False
+        >>> something # skip if not HAVE_AMODULE
+        0
+        >>> something # skip if HAVE_BMODULE
+        0
+
+    This decorator will evaluate the expression after ``skip if``.  If this
+    evaluates to True, then the comment is replaced by ``# doctest: +SKIP``. If
+    False, then the comment is just removed. The expression is evaluated in the
+    ``globals`` scope of `func`.
+
+    For example, if the module global ``HAVE_AMODULE`` is False, and module
+    global ``HAVE_BMODULE`` is False, the returned function will have docstring::
+
+        >>> something # doctest: +SKIP
+        >>> something + else # doctest: +SKIP
+        >>> something # doctest: +SKIP
+
+    """
+    lines = func.__doc__.split('\n')
+    new_lines = []
+    for line in lines:
+        match = SKIP_RE.match(line)
+        if match is None:
+            new_lines.append(line)
+            continue
+        code, space, expr = match.groups()
+
+        try:
+            # Works as a function decorator
+            if eval(expr, func.__globals__):
+                code = code + space + "# doctest: +SKIP"
+        except AttributeError:
+            # Works as a class decorator
+            if eval(expr, func.__init__.__globals__):
+                code = code + space + "# doctest: +SKIP"
+
+        new_lines.append(code)
+    func.__doc__ = "\n".join(new_lines)
+    return func
+
+
+def roundtrip(image, plugin, suffix):
+    """Save and read an image using a specified plugin"""
+    if '.' not in suffix:
+        suffix = '.' + suffix
+    with NamedTemporaryFile(suffix=suffix, delete=False) as temp_file:
+        fname = temp_file.name
+    io.imsave(fname, image, plugin=plugin)
+    new = io.imread(fname, plugin=plugin)
+    try:
+        os.remove(fname)
+    except Exception:
+        pass
+    return new
+
+
+def color_check(plugin, fmt='png'):
+    """Check roundtrip behavior for color images.
+
+    All major input types should be handled as ubytes and read
+    back correctly.
+    """
+    img = img_as_ubyte(data.chelsea())
+    r1 = roundtrip(img, plugin, fmt)
+    testing.assert_allclose(img, r1)
+
+    img2 = img > 128
+    r2 = roundtrip(img2, plugin, fmt)
+    testing.assert_allclose(img2, r2.astype(bool))
+
+    img3 = img_as_float(img)
+    r3 = roundtrip(img3, plugin, fmt)
+    testing.assert_allclose(r3, img)
+
+    img4 = img_as_int(img)
+    if fmt.lower() in (('tif', 'tiff')):
+        img4 -= 100
+        r4 = roundtrip(img4, plugin, fmt)
+        testing.assert_allclose(r4, img4)
+    else:
+        r4 = roundtrip(img4, plugin, fmt)
+        testing.assert_allclose(r4, img_as_ubyte(img4))
+
+    img5 = img_as_uint(img)
+    r5 = roundtrip(img5, plugin, fmt)
+    testing.assert_allclose(r5, img)
+
+
+def mono_check(plugin, fmt='png'):
+    """Check the roundtrip behavior for images that support most types.
+
+    All major input types should be handled.
+    """
+
+    img = img_as_ubyte(data.moon())
+    r1 = roundtrip(img, plugin, fmt)
+    testing.assert_allclose(img, r1)
+
+    img2 = img > 128
+    r2 = roundtrip(img2, plugin, fmt)
+    testing.assert_allclose(img2, r2.astype(bool))
+
+    img3 = img_as_float(img)
+    r3 = roundtrip(img3, plugin, fmt)
+    if r3.dtype.kind == 'f':
+        testing.assert_allclose(img3, r3)
+    else:
+        testing.assert_allclose(r3, img_as_uint(img))
+
+    img4 = img_as_int(img)
+    if fmt.lower() in (('tif', 'tiff')):
+        img4 -= 100
+        r4 = roundtrip(img4, plugin, fmt)
+        testing.assert_allclose(r4, img4)
+    else:
+        r4 = roundtrip(img4, plugin, fmt)
+        testing.assert_allclose(r4, img_as_uint(img4))
+
+    img5 = img_as_uint(img)
+    r5 = roundtrip(img5, plugin, fmt)
+    testing.assert_allclose(r5, img5)
+
+
+def fetch(data_filename):
+    """Attempt to fetch data, but if unavailable, skip the tests."""
+    try:
+        return _fetch(data_filename)
+    except (ConnectionError, ModuleNotFoundError):
+        pytest.skip(f'Unable to download {data_filename}', allow_module_level=True)
+
+
+def run_in_parallel(num_threads=2, warnings_matching=None):
+    """Decorator to run the same function multiple times in parallel.
+
+    This decorator is useful to ensure that separate threads execute
+    concurrently and correctly while releasing the GIL.
+
+    Parameters
+    ----------
+    num_threads : int, optional
+        The number of times the function is run in parallel.
+
+    warnings_matching: list or None
+        This parameter is passed on to `expected_warnings` so as not to have
+        race conditions with the warnings filters. A single
+        `expected_warnings` context manager is used for all threads.
+        If None, then no warnings are checked.
+
+    """
+
+    assert num_threads > 0
+
+    def wrapper(func):
+        @functools.wraps(func)
+        def inner(*args, **kwargs):
+            with expected_warnings(warnings_matching):
+                threads = []
+                for i in range(num_threads - 1):
+                    thread = threading.Thread(target=func, args=args, kwargs=kwargs)
+                    threads.append(thread)
+                for thread in threads:
+                    thread.start()
+
+                func(*args, **kwargs)
+
+                for thread in threads:
+                    thread.join()
+
+        return inner
+
+    return wrapper
+
+
+def assert_stacklevel(warnings, *, offset=-1):
+    """Assert correct stacklevel of captured warnings.
+
+    When scikit-image raises warnings, the stacklevel should ideally be set
+    so that the origin of the warnings will point to the public function
+    that was called by the user and not necessarily the very place where the
+    warnings were emitted (which may be inside of some internal function).
+    This utility function helps with checking that
+    the stacklevel was set correctly on warnings captured by `pytest.warns`.
+
+    Parameters
+    ----------
+    warnings : collections.abc.Iterable[warning.WarningMessage]
+        Warnings that were captured by `pytest.warns`.
+    offset : int, optional
+        Offset from the line this function is called to the line were the
+        warning is supposed to originate from. For multiline calls, the
+        first line is relevant. Defaults to -1 which corresponds to the line
+        right above the one where this function is called.
+
+    Raises
+    ------
+    AssertionError
+        If a warning in `warnings` does not match the expected line number or
+        file name.
+
+    Examples
+    --------
+    >>> def test_something():
+    ...     with pytest.warns(UserWarning, match="some message") as record:
+    ...         something_raising_a_warning()
+    ...     assert_stacklevel(record)
+    ...
+    >>> def test_another_thing():
+    ...     with pytest.warns(UserWarning, match="some message") as record:
+    ...         iam_raising_many_warnings(
+    ...             "A long argument that forces the call to wrap."
+    ...         )
+    ...     assert_stacklevel(record, offset=-3)
+    """
+    frame = inspect.stack()[1].frame  # 0 is current frame, 1 is outer frame
+    line_number = frame.f_lineno + offset
+    filename = frame.f_code.co_filename
+    expected = f"{filename}:{line_number}"
+    for warning in warnings:
+        actual = f"{warning.filename}:{warning.lineno}"
+        assert actual == expected, f"{actual} != {expected}"

skimage/_shared/tests/test_coord.py

@@ -0,0 +1,91 @@
+import time
+
+import numpy as np
+import pytest
+from scipy.spatial.distance import pdist, minkowski
+
+from skimage._shared.coord import ensure_spacing
+
+
+@pytest.mark.parametrize("p", [1, 2, np.inf])
+@pytest.mark.parametrize("size", [30, 50, None])
+def test_ensure_spacing_trivial(p, size):
+    # --- Empty input
+    assert ensure_spacing([], p_norm=p) == []
+
+    # --- A unique point
+    coord = np.random.randn(1, 2)
+    assert np.array_equal(coord, ensure_spacing(coord, p_norm=p, min_split_size=size))
+
+    # --- Verified spacing
+    coord = np.random.randn(100, 2)
+
+    # --- 0 spacing
+    assert np.array_equal(
+        coord, ensure_spacing(coord, spacing=0, p_norm=p, min_split_size=size)
+    )
+
+    # Spacing is chosen to be half the minimum distance
+    spacing = pdist(coord, metric=minkowski, p=p).min() * 0.5
+
+    out = ensure_spacing(coord, spacing=spacing, p_norm=p, min_split_size=size)
+
+    assert np.array_equal(coord, out)
+
+
+@pytest.mark.parametrize("ndim", [1, 2, 3, 4, 5])
+@pytest.mark.parametrize("size", [2, 10, None])
+def test_ensure_spacing_nD(ndim, size):
+    coord = np.ones((5, ndim))
+
+    expected = np.ones((1, ndim))
+
+    assert np.array_equal(ensure_spacing(coord, min_split_size=size), expected)
+
+
+@pytest.mark.parametrize("p", [1, 2, np.inf])
+@pytest.mark.parametrize("size", [50, 100, None])
+def test_ensure_spacing_batch_processing(p, size):
+    coord = np.random.randn(100, 2)
+
+    # --- Consider the average distance btween the point as spacing
+    spacing = np.median(pdist(coord, metric=minkowski, p=p))
+
+    expected = ensure_spacing(coord, spacing=spacing, p_norm=p)
+
+    assert np.array_equal(
+        ensure_spacing(coord, spacing=spacing, p_norm=p, min_split_size=size), expected
+    )
+
+
+def test_max_batch_size():
+    """Small batches are slow, large batches -> large allocations -> also slow.
+
+    https://github.com/scikit-image/scikit-image/pull/6035#discussion_r751518691
+    """
+    coords = np.random.randint(low=0, high=1848, size=(40000, 2))
+    tstart = time.time()
+    ensure_spacing(coords, spacing=100, min_split_size=50, max_split_size=2000)
+    dur1 = time.time() - tstart
+
+    tstart = time.time()
+    ensure_spacing(coords, spacing=100, min_split_size=50, max_split_size=20000)
+    dur2 = time.time() - tstart
+
+    # Originally checked dur1 < dur2 to assert that the default batch size was
+    # faster than a much larger batch size. However, on rare occasion a CI test
+    # case would fail with dur1 ~5% larger than dur2. To be more robust to
+    # variable load or differences across architectures, we relax this here.
+    assert dur1 < 1.33 * dur2
+
+
+@pytest.mark.parametrize("p", [1, 2, np.inf])
+@pytest.mark.parametrize("size", [30, 50, None])
+def test_ensure_spacing_p_norm(p, size):
+    coord = np.random.randn(100, 2)
+
+    # --- Consider the average distance btween the point as spacing
+    spacing = np.median(pdist(coord, metric=minkowski, p=p))
+    out = ensure_spacing(coord, spacing=spacing, p_norm=p, min_split_size=size)
+
+    assert pdist(out, metric=minkowski, p=p).min() > spacing

skimage/_shared/tests/test_dtype.py

@@ -0,0 +1,14 @@
+import numpy as np
+import pytest
+
+
+from ..dtype import numeric_dtype_min_max, numeric_types
+
+
+class Test_numeric_dtype_min_max:
+    @pytest.mark.parametrize("dtype", numeric_types)
+    def test_all_numeric_types(self, dtype):
+        min_, max_ = numeric_dtype_min_max(dtype)
+        assert np.isscalar(min_)
+        assert np.isscalar(max_)
+        assert min_ < max_