BUG: ValueError for input [1, 1] in morphology.local_maxima when [0, 0] is expected #3261
Description
The error can be reproduced with all development versions of scikit-image that include 912d9d2 and the following snippet:
>>> from skimage.morphology import local_maxima
>>> local_maxima([1, 1], allow_borders=False)This raises
----> 1 local_maxima([1, 1], allow_borders=False)
~/Projects/skimage/skimage/morphology/extrema.py in local_maxima(image, selem, connectivity, indices, allow_borders)
441 selem = _resolve_neighborhood(selem, connectivity, image.ndim)
442 neighbor_offsets = _offsets_to_raveled_neighbors(
--> 443 image.shape, selem, center=((1,) * image.ndim))
444
445 # Array of flags used to store the state of each pixel during evaluation.
~/Projects/skimage/skimage/morphology/watershed.py in _offsets_to_raveled_neighbors(image_shape, structure, center)
148 structure[tuple(center)] = 0 # Ignore the center; it's not a neighbor
149 connection_indices = np.transpose(np.nonzero(structure))
--> 150 offsets = (np.ravel_multi_index(connection_indices.T, image_shape) -
151 np.ravel_multi_index(center, image_shape))
152 squared_distances = np.sum((connection_indices - center) ** 2, axis=1)
ValueError: invalid entry in coordinates array
but should return array([0, 0]).
Until me or someone else finds the time to implement a fix and regression test this issue acts as a reminder.
Discovered in #3255.