def test_intensity_remap_values():
trf = slt.IntensityRemap(p=1)
img = np.arange(0, 256, 1, dtype=np.uint8).reshape((16, 16, 1))
dc = slc.DataContainer(img, "I")
out = trf(dc).data[0]
# Mapping is applied correctly
img_expected = trf.state_dict["LUT"].reshape((16, 16, 1))
np.testing.assert_array_equal(out, img_expected)
# Mapping has a positive trendline
assert np.sum(np.diff(out.astype(np.float).ravel())) > 0
# Higher kernel size yields more monotonic mapping
trf_noisy = slt.IntensityRemap(p=1, kernel_size=1)
trf_low_pass = slt.IntensityRemap(p=1, kernel_size=5)
out_noisy = trf_noisy(dc).data[0].astype(np.float)
out_low_pass = trf_low_pass(dc).data[0].astype(np.float)
std_noisy = np.std(np.diff(out_noisy.ravel()))
std_low_pass = np.std(np.diff(out_low_pass.ravel()))
assert std_low_pass < std_noisy
def test_intensity_remap_dtypes(dtype, expected):
trf = slt.IntensityRemap(p=1)
dc = slc.DataContainer(img_3x3().astype(dtype), "I")
with expected:
trf(dc)
def test_intensity_remap_channels(img, expected):
trf = slt.IntensityRemap(p=1)
dc = slc.DataContainer(img, "I")
with expected:
trf(dc)