search

zaccharieramzi / tfkbnufft

A robust, easy-to-deploy non-uniform Fast Fourier Transform in TensorFlow.
MIT License
30 stars 5 forks source link

Allow dynamically shaped input #4

Closed zaccharieramzi closed 4 years ago

zaccharieramzi commented 4 years ago

To handle fastMRI data, tfkbnufft needs to be able to handle dyanmically shaped input.

Currently, this is not doable because you have to pass in the image shape to KbNufftModule.

What we would need is to see what is affected by potentially having the shape only at inference time. Of course there should still be a way to specify the shape beforehand for a computation speedup.

zaccharieramzi commented 4 years ago

This would actually mean re-doing all the numpy utils from nfft.utils. This is actually quite involved and might be source of a lot of errors.

Maybe we need to evaluate a different route to use for dyanmically shaped input, like padding. It's actually not that stupid in this case because we can easily specify the coordinates to take into account.