Chris Markiewicz
Chris Markiewicz
What about `load_niimg`? [This line](https://github.com/nilearn/nilearn/blob/ee1f25a908d8357b4fba84dab1d2b7018e9f5e04/nilearn/_utils/niimg.py#L127) looks like a potential memory grab.
And this is on an uncompressed file?
Actually, can you just: ```Python print(img.header) ```
Oh, right. I forgot that nibabel destroys the slope/intercept information. Anyway, given that it's int16, that almost certainly means it's scaled, and therefore being upcast to a `float64` in `img.dataobj.__array__()`....
Hi @synchon! Thanks for your interest. The pairwise transformations have been part of the vision for templateflow, but have fallen by the wayside with nobody really making it a priority...
+1 on this. @oesteban any reason not to?
Is this related to files no longer being available on S3 or should I open a separate issue?
I can open a new issue tomorrow. Pretty much all of OASIS is gone. Haven't made a census, but CI is failing on niworkflows at least.
I have no objections to GallOP being under con/, and can't immediately think of a more appropriate alternative.
That would be a good way to test out alternatives before committing to a change.