Describe your issue.
In SciPy 1.11, axes support was added to a number of ndimage filters. While porting this to CuPy I found that for minimum_filter and maximum_filter when all three of the following criteria are true, there is a bug in origin argument handling:
len(axes) < ndim- a non-separable
footprint array is provided
- origin is provided as a tuple
I have a fix for the issue and will submit a small MR for it.
Reproducing Code Example
import numpy as np
from scipy.ndimage import maximum_filter
img = np.arange((5 * 6 * 7), dtype=np.float64).reshape(5, 6, 7)
footprint = np.ones((3, 5), dtype=bool)
footprint[1, 0] = 0
out = maximum_filter(img, footprint=footprint, axes=(-2, -1), origin=(0, 1))
Error message
RuntimeError Traceback (most recent call last)
Cell In[3], line 7
5 footprint = np.ones((3, 5), dtype=bool)
6 footprint[1, 0] = 0
----> 7 out = maximum_filter(img, footprint=footprint, axes=(-2, -1), origin=(0, 1))
File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_filters.py:1331, in maximum_filter(input, size, footprint, output, mode, cval, origin, axes)
1287 @_ni_docstrings.docfiller
1288 def maximum_filter(input, size=None, footprint=None, output=None,
1289 mode="reflect", cval=0.0, origin=0, *, axes=None):
1290 """Calculate a multidimensional maximum filter.
1291
1292 Parameters
(...)
1329 >>> plt.show()
1330 """
-> 1331 return _min_or_max_filter(input, size, footprint, None, output, mode,
1332 cval, origin, 0, axes)
File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_filters.py:1200, in _min_or_max_filter(input, size, footprint, structure, output, mode, cval, origin, minimum, axes)
1198 output[...] = input[...]
1199 else:
-> 1200 origins = _ni_support._normalize_sequence(origin, input.ndim)
1201 if num_axes < input.ndim:
1202 if footprint.ndim != num_axes:
File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_ni_support.py:68, in _normalize_sequence(input, rank)
66 if len(normalized) != rank:
67 err = "sequence argument must have length equal to input rank"
---> 68 raise RuntimeError(err)
69 else:
70 normalized = [input] * rank
RuntimeError: sequence argument must have length equal to input rank
### SciPy/NumPy/Python version and system information
```shell
1.11.1 1.26.4 sys.version_info(major=3, minor=10, micro=12, releaselevel='final', serial=0)
Describe your issue.
In SciPy 1.11, axes support was added to a number of ndimage filters. While porting this to CuPy I found that for
minimum_filterandmaximum_filterwhen all three of the following criteria are true, there is a bug inoriginargument handling:len(axes) < ndimfootprintarray is providedI have a fix for the issue and will submit a small MR for it.
Reproducing Code Example
Error message
RuntimeError Traceback (most recent call last) Cell In[3], line 7 5 footprint = np.ones((3, 5), dtype=bool) 6 footprint[1, 0] = 0 ----> 7 out = maximum_filter(img, footprint=footprint, axes=(-2, -1), origin=(0, 1)) File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_filters.py:1331, in maximum_filter(input, size, footprint, output, mode, cval, origin, axes) 1287 @_ni_docstrings.docfiller 1288 def maximum_filter(input, size=None, footprint=None, output=None, 1289 mode="reflect", cval=0.0, origin=0, *, axes=None): 1290 """Calculate a multidimensional maximum filter. 1291 1292 Parameters (...) 1329 >>> plt.show() 1330 """ -> 1331 return _min_or_max_filter(input, size, footprint, None, output, mode, 1332 cval, origin, 0, axes) File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_filters.py:1200, in _min_or_max_filter(input, size, footprint, structure, output, mode, cval, origin, minimum, axes) 1198 output[...] = input[...] 1199 else: -> 1200 origins = _ni_support._normalize_sequence(origin, input.ndim) 1201 if num_axes < input.ndim: 1202 if footprint.ndim != num_axes: File ~/mambaforge/envs/cucim310/lib/python3.10/site-packages/scipy/ndimage/_ni_support.py:68, in _normalize_sequence(input, rank) 66 if len(normalized) != rank: 67 err = "sequence argument must have length equal to input rank" ---> 68 raise RuntimeError(err) 69 else: 70 normalized = [input] * rank RuntimeError: sequence argument must have length equal to input rank