Is it possible to use *cfl files with this library? about torchkbnufft HOT 3 CLOSED

Hello @mscipio, could you describe what your dimensions mean? To me it looks like a 20-coil dataset. As you can see from the KbNufftAdjoint documentation the input size should be (N, C) + klength, where N is the batch size and C is the number of coils, so I think your inputs should be [1, 20, 44636800] for the data and [3, 44636800] for the trajectory.

Also from the documentation, the units of the k-space should be in radians/voxel, so there may be some conversions you need for the trajectory before inputting it into the operator.

Lastly, I'll say that the package can be a little slow with 3D trajectories, as stated in the performance notes. If there are any transformations you can make to a 2D problem (e.g., a stack-of-stars radial trajectory), I highly recommend using those to improve computation speed.

from torchkbnufft.

Thank you very much for those pointers, @mmuckley
I will make sure to double-check I am doing everything right and then I'll share a (hopefully successful update).

I think I also have a 'trick' I can use to turn the 3D problem into a 2D problem, so I will try that as well.

from torchkbnufft.

Closing as there has been no follow-up. Feel free to reopen if necessary.

from torchkbnufft.