michaelbratsch
michaelbratsch
I agree that dispatching is possible but what is the point of keeping `binary_dilation/closing`? It has fewer `modes` than `dilation/closing` and the original purpose was to have a faster interface...
I tried the dispatching approach and then stumbled upon the same issues as being noted in https://github.com/scikit-image/scikit-image/issues/7238, `dilation` and `binary_dilation` have some inconsistencies in the way they operate, see >...
I can reproduce the issue in the description via: ```python import sys print("Before: ", [x for x in sys.modules if "matplotlib" in x]) from skimage.morphology import flood_fill print("After: ", [x...
If there would be a folder like `skimage/_build_utils` but for testing then we could move the boolean `has_mpl` there so it is only being imported by test code. Do we...
The unit test is currently known to be failing for 32-bit environment: ``` @xfail( condition=arch32, reason=( 'Known test failure on 32-bit platforms. See links for ' 'details: ' 'https://github.com/scikit-image/scikit-image/issues/3091 '...
The unit test was probably intending to trigger this `return False`: https://github.com/scikit-image/scikit-image/blob/a160f384523ac70173d069954b7084bee79fb68e/skimage/measure/fit.py#L494-L495 because for three given points https://github.com/scikit-image/scikit-image/blob/a160f384523ac70173d069954b7084bee79fb68e/skimage/measure/fit.py#L480 `M` is exactly zero if the matrix D2 is nonsingular: ``` S1...
Unrelated to the above, the `estimate` of `EllipseModel` is also unstable in the sense that I can provoke with the input points `[[0, 0], [2, 3], [4, 6]]` the following...
> Should we be using `eigh` instead of `eig`? I'm a bit nervous about checking for _any_ complex component present; not sure how sensitive that would be. I wrote in...
If I switch from numpy 2.0.1 (using accelerate) to numpy 1.26.4 (using openblas64), the `test_ellipse_model_estimate_failers` is pass but `test_ellipse_parameter_stability` is failing with > phi = 0.5 * np.arctan((2.0 * b)...
@jakublipinski I think in your case, increasing `min_samples` of `ransac` to 5 (or more) should help to avoid the instability because the ellipse estimation is not underdetermined anymore. If the...