# MRI-NUFFT fails to detect available backends

mind-inria / mri-nufft

Doing non-Cartesian MR Imaging has never been so easy.

https://mind-inria.github.io/mri-nufft/

BSD 3-Clause "New" or "Revised" License

54 stars 10 forks source link

# MRI-NUFFT fails to detect available backends #178

Closed mcencini closed 2 months ago

mcencini commented 3 months ago

Hi all, thanks for this awesome package! I have noticed that the package seems to fail to correctly detect available backends. In my case, I only installed torchkbnufft backend, but when I run the tests I obtain this:

backends: avail select stacked: True True stacked-cufinufft: True True bart: False False cufinufft: True True finufft: REF REF gpunufft: False False pynfft: True True numpy: True True pynufft-cpu: False False sigpy: False False tensorflow: False False torchkbnufft-cpu: True True torchkbnufft-gpu: True True

I would expect pynfft and cufinufft / finufft to return False (and, since finufft is the reference backend, to display an error). Thanks!

Best wishes, Matteo

paquiteau commented 3 months ago

Hi, thanks for opening the issue.

Could you provide more information on how you install mri-nufft , also what packages are installed in your environment (pip freeze should gives us this information).
With which command did you run the test ? you can filter backends to test using the --backend and --ref argument like so:
```
pytest --backend torchkbnufft --ref finufft 
```
if those arguments are not specified, all available backends will be tested.

mcencini commented 3 months ago

Hi, sorry for the late reply. I installed mri-nufft on a conda environment (Python 3.12.4). Test seems to actually skip the uninstalled PyNFFT backend even if it is displayed as detected and selected. I will try to install again on a fresh environment and let you know if the problem persists. Thanks!