ValueError: need at least one array to concatenate with upgraded Numpy

Question

I'm bringing dependencies up to date in an old project of mine and recently ran into an issue when using RandomWalker as follows:

    import numpy as np
    from skimage import color as skc
    from skimage.segmentation import random_walker


    def my_func(img):
        img = skc.rgb2gray(img)

        markers = np.zeros_like(img, dtype=np.uint)
        markers[img < (img.mean() - img.std()) / 2.0] = 1
        markers[img > (img.mean() + img.std()) / 2.0] = 2

        try:
            rw = random_walker(img, markers) # ValueError: need at least one array to concatenate
        except RuntimeError:
            ...

The dependencies I previously got and that are currently working just fine are (if there is any other version required that can help troubleshooting just ask):

# Python 3.6
numpy==1.14.5
scikit-image==0.13.0
scikit-learn==0.19.2

And upgraded ones where the ValueError: need at least one array to concatenate is thrown are:

# Python 3.9
numpy==1.21.0
scikit-image==0.19.1
scikit-learn==1.1.2

Example for testing - install previously mentioned dependencies from requirements.txt , with the Python versions mentioned:

import numpy as np
from skimage.segmentation import random_walker
from skimage import color as skc


img_array = np.array(
    [
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
        [
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
            [255, 255, 255],
        ],
    ]
).astype(np.uint8)


def walker_texas_ranger():
    image = skc.rgb2gray(img_array)

    markers = np.zeros_like(image, dtype=np.uint)
    markers[image < image.mean() - image.std() / 2.0] = 1
    markers[image > image.mean() + image.std() / 2.0] = 2

    try:
        rw = random_walker(image, markers)
    except RuntimeError:
        return None

    return rw


if __name__ == '__main__':
    walker_texas_ranger()

Sample execution stack trace running latest dependencies with Python 3.9 :

  File "/Users/me/examples/main.py", line 188, in walker_texas_ranger
    rw = random_walker(image, markers)
  File "/usr/local/lib/python3.9/site-packages/skimage/_shared/utils.py", line 338, in fixed_func
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.9/site-packages/skimage/_shared/utils.py", line 293, in fixed_func
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.9/site-packages/skimage/segmentation/random_walker_segmentation.py", line 490, in random_walker
    lap_sparse, B = _build_linear_system(data, spacing, labels, nlabels, mask,
  File "/usr/local/lib/python3.9/site-packages/skimage/segmentation/random_walker_segmentation.py", line 157, in _build_linear_system
    seeds_mask = sparse.csc_matrix(np.hstack(
  File "<__array_function__ internals>", line 5, in hstack
  File "/usr/local/lib/python3.9/site-packages/numpy/core/shape_base.py", line 345, in hstack
    return _nx.concatenate(arrs, 1)
  File "<__array_function__ internals>", line 5, in concatenate
ValueError: need at least one array to concatenate

Any clue on what has changed so that it is no longer working?

Answer 1

The issue is here:

    markers = np.zeros_like(image, dtype=np.uint)
    markers[image < image.mean() - image.std() / 2.0] = 1
    markers[image > image.mean() + image.std() / 2.0] = 2

Your image only contains values of 255. So image.mean() - image.std() / 2.0 is 255, and image.mean() + image.std() / 2.0 is also 255, and no image pixels are < or > these values. Therefore, your markers array contains all zeros and there are no walkers, which causes the error. I'm not sure what the old behaviour would have been.

Here's a more minimal reproducer causing the same error:

import numpy as np
from skimage import segmentation

image = np.random.random((5, 5))
labels = np.zeros((5, 5), dtype=int)
segmented = segmentation.random_walker(image, labels)

Either way, the error raised should be less obscure, so I've created an issue in the scikit-image repo: scikit-image/scikit-image#6561 .