Silent failing of CPU 3D median filter when array of incorrect data type is passed

[yousefmoazzam]

When testing for tomopy/tomopy#594, it was discovered that passing a python numpy array of values whose data type is not supported by the CPU median filter module https://github.com/dkazanc/larix/blob/master/src/Core/CPU_modules/MedianFilt_core.c causes the python wrapper function to simply return None.

For example, when the input array has dtype=np.float64 (64-bit float arrays aren't supported by Larix's CPU 3D median filter), there is no error and it instead silently fails:

>>> import numpy as np
>>> from larix.methods.misc import MEDIAN_FILT
>>> float64_data = np.ones(shape=(100, 100, 100), dtype=np.float64)
>>> float64_res = MEDIAN_FILT(float64_data, 3, 4)
>>> float64_res
>>> type(float64_res)
<class 'NoneType'>

but when the input array has dtype=np.uint16 (16-bit unsigned integer arrays are supported by Larix's CPU 3D median filter), it works as expected:

>>> uint16_data = np.ones(shape=(100, 100, 100), dtype=np.uint16)
>>> uint16_res = MEDIAN_FILT(uint16_data, 3, 4)
>>> uint16_res
array([[[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]],

       [[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]],

       [[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]],

       ...,

       [[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]],

       [[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]],

       [[1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        ...,
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1],
        [1, 1, 1, ..., 1, 1, 1]]], dtype=uint16)
>>> type(uint16_res)
<class 'numpy.ndarray'>

This is possibly something in the python wrapper that needs to be amended, rather than the median filter C code itself?

[namannimmo10]

MEDIAN_FILT is here:

larix/src/Python/src/cpu_misc.pyx

Lines 82 to 92 in 06f3952

	def MEDIAN_FILT(Input, radius, ncores=0):
	input_type = Input.dtype
	if ((Input.ndim == 2) and (input_type == 'float32')):
	return MEDIAN_FILT_float32_2D(Input, radius, ncores)
	elif ((Input.ndim == 2) and (input_type == 'uint16')):
	return MEDIAN_FILT_uint16_2D(Input, radius, ncores)
	elif ((Input.ndim == 3) and (input_type == 'float32')):
	return MEDIAN_FILT_float32_3D(Input, radius, ncores)
	elif ((Input.ndim == 3) and (input_type == 'uint16')):
	return MEDIAN_FILT_uint16_3D(Input, radius, ncores)