add test cases

python/cucim/src/cucim/skimage/_vendored/tests/test_morphology.py

@@ -0,0 +1,131 @@
+"""Testing axes support that is not currently present in CuPy/SciPy."""
+
+import cupy as cp
+import numpy as np
+import pytest
+import scipy.ndimage as ndi_cpu
+
+from cucim.skimage._vendored import ndimage as ndi
+
+
+@pytest.mark.parametrize("border_value", [0, 1])
+@pytest.mark.parametrize("origin", [(0, 0), (-1, 0)])
+@pytest.mark.parametrize("expand_axis", [0, 1, 2])
+@pytest.mark.parametrize(
+    "func",
+    [
+        "binary_erosion",
+        "binary_dilation",
+        "binary_opening",
+        "binary_closing",
+        "binary_hit_or_miss",
+        "binary_propagation",
+        "binary_fill_holes",
+    ],
+)
+def test_binary_axes(func, expand_axis, origin, border_value):
+    struct = [[0, 1, 0], [1, 1, 1], [0, 1, 0]]
+
+    data = np.array(
+        [
+            [0, 0, 0, 1, 0, 0, 0],
+            [0, 0, 0, 1, 0, 0, 0],
+            [0, 0, 1, 1, 0, 1, 0],
+            [0, 1, 0, 1, 1, 0, 1],
+            [0, 1, 1, 1, 1, 1, 0],
+            [0, 0, 1, 1, 0, 0, 0],
+            [0, 0, 0, 1, 0, 0, 0],
+        ],
+        bool,
+    )
+    if func == "binary_hit_or_miss":
+        kwargs = dict(origin1=origin, origin2=origin)
+    else:
+        kwargs = dict(origin=origin)
+    border_supported = func not in ["binary_hit_or_miss", "binary_fill_holes"]
+    if border_supported:
+        kwargs["border_value"] = border_value
+    elif border_value != 0:
+        pytest.skip("border_value !=0 unsupported by this function")
+
+    scipy_func = getattr(ndi_cpu, func)
+    expected = scipy_func(data, struct, **kwargs)
+
+    # copy data and expected results to GPU
+    data = cp.asarray(data)
+    expected = cp.asarray(expected)
+    struct = cp.asarray(struct)
+
+    # replicate data and expected result along a new axis
+    n_reps = 5
+    expected = cp.stack([expected] * n_reps, axis=expand_axis)
+    data = cp.stack([data] * n_reps, axis=expand_axis)
+
+    # filter all axes except expand_axis
+    axes = [0, 1, 2]
+    axes.remove(expand_axis)
+    out = cp.zeros(data.shape, bool)
+    cucim_func = getattr(ndi, func)
+    cucim_func(data, struct, output=out, axes=axes, **kwargs)
+    cp.testing.assert_array_almost_equal(out, expected)
+
+
+@pytest.mark.parametrize("origin", [(0, 0), (-1, 0)])
+@pytest.mark.parametrize("expand_axis", [0, 1, 2])
+@pytest.mark.parametrize("mode", ["reflect", "constant"])
+@pytest.mark.parametrize("footprint_mode", ["size", "footprint", "structure"])
+@pytest.mark.parametrize(
+    "func",
+    [
+        "grey_erosion",
+        "grey_dilation",
+        "grey_opening",
+        "grey_closing",
+        "morphological_laplace",
+        "morphological_gradient",
+        "white_tophat",
+        "black_tophat",
+    ],
+)
+def test_grey_axes(func, expand_axis, origin, footprint_mode, mode):
+    data = np.array(
+        [
+            [0, 0, 0, 1, 0, 0, 0],
+            [0, 0, 0, 4, 0, 0, 0],
+            [0, 0, 2, 1, 0, 2, 0],
+            [0, 3, 0, 6, 5, 0, 1],
+            [0, 4, 5, 3, 3, 4, 0],
+            [0, 0, 9, 3, 0, 0, 0],
+            [0, 0, 0, 2, 0, 0, 0],
+        ]
+    )
+    kwargs = dict(origin=origin, mode=mode)
+    if footprint_mode == "size":
+        kwargs["size"] = (2, 3)
+    else:
+        kwargs["footprint"] = np.asarray([[1, 0, 1], [1, 1, 0]])
+    if footprint_mode == "structure":
+        kwargs["structure"] = np.ones_like(kwargs["footprint"])
+    scipy_func = getattr(ndi_cpu, func)
+    expected = scipy_func(data, **kwargs)
+
+    # copy data and expected results to GPU
+    data = cp.asarray(data)
+    expected = cp.asarray(expected)
+    if "footprint" in kwargs:
+        kwargs["footprint"] = cp.asarray(kwargs["footprint"])
+    if "structure" in kwargs:
+        kwargs["structure"] = cp.asarray(kwargs["structure"])
+
+    # replicate data and expected result along a new axis
+    n_reps = 5
+    expected = cp.stack([expected] * n_reps, axis=expand_axis)
+    data = cp.stack([data] * n_reps, axis=expand_axis)
+
+    # filter all axes except expand_axis
+    axes = [0, 1, 2]
+    axes.remove(expand_axis)
+    out = cp.zeros(expected.shape, dtype=expected.dtype)
+    cucim_func = getattr(ndi, func)
+    cucim_func(data, output=out, axes=axes, **kwargs)
+    cp.testing.assert_array_almost_equal(out, expected)