Constantin Pape

> I agree with Constantin's proposal and I would rename "Region" to "Superpixel" in this already awesomely optimized table. Good point, that's more obvious to users in this context.

Thanks for raising the point about the additional probability maps @stuarteberg, I think if we have a very simple menu as in https://github.com/ilastik/ilastik/issues/2546#issuecomment-1041303746 we could add all probability maps, given...

A few comments from my side: - I assume that "boundary" and "prob-0" are actually different, i.e. that you are not repeating the probability input that is used as boundary...

> For other features, I'm not sure when I would not select all. Or, when would you select skewness and not kurtosis? Does "count" alone make any sense? I agree...

Sounds good to me.

If I understand your code snippet correctly, you want to be able to pass a dict `attributes` to `add_node`, correct? How would you use these in the nodes? Access them...

Regarding the first error: `scipy.imsave` is deprecated and not included in more recent scipy versions anymore. We fixed this in #177, unfortunately this is not reflected in the conda-forge version...

I could not reproduce this on current master (also using `tensorboardx 1.8`). I have updated the conda-forge version, but the new version will only be online in a few hours,...

Do you need this on pypi? The recommended way to install is via conda-forge.

Yeah, I agree. @nasimrahaman uploaded it, so I think he has to do that.