scisprints / 2018_05_sklearn_skimage_dask

BSD 3-Clause "New" or "Revised" License
6 stars 2 forks source link

n-Dimensional transformations for image processing #6

Open kne42 opened 6 years ago

kne42 commented 6 years ago

I have been considering expanding the transforms module of scikit-image to support n-dimensional operations. This would provide more utility and lay the groundwork to start expanding the rest of the library to work in nD as well.

Essential matrices will be an issue however since they make use of the wedge product, which is not implemented in numpy.

kne42 commented 6 years ago

cc @jni

jni commented 6 years ago

@kne42 you should practice your math lecturing skills before the sprint, so you can explain this gibberish to me. 😂

Incidentally, it is ok for a subset of transforms to only support a subset of dimensionalities, as long as it is all well-documented. The principal goal is to make sure our coordinates match NumPy axes for the whole transform module.

jakirkham commented 6 years ago

Have you played with atop at all?

kne42 commented 6 years ago

@jakirkham No, I have not. It looks very useful though! I should probably familiarize myself with dask before the sprint lol...

jakirkham commented 6 years ago

Should probably add an einsum implementation was added very recently as well. Maybe that’s even a better option depending on what’s required for these operations.

kne42 commented 6 years ago

The einsum function should do nicely for wedge product computation; thank you!