1D line extraction from 2D images

[McHaillet]

I worked on the 2D to 1D Fourier slice extraction, which is functional (I tried it for tttsa, see teamtomo/tttsa#19).

One thing that I haven't figured out is a 90 degrees addition to the in plane rotations for the rotation matrices for this function, compared to the affine transform I used before (link to the line of code). One of the two is interpreting it wrongly.

The dft_center also needed a fix to run on GPU, I can also put that in a separate PR if you are not sure yet about this one and want that bug fix in.

I have not added the inverse operation of 1D to 2D slice insertion. Should that be added here as well?

[alisterburt]

awesome - really clean (but I would say that... it mirrors my own implementation 😆)

few minor notes but happy for this to be merged when you're ready

re: the 90 degree rotation I can't see anything obvious - the difference I would expect between conventions/implementations would be an inversion of rotation angle rather than a 90 degree shift, is this actually what is happening? does a zero degree projection produce an image projected along y as expected? what about 90 and x?

[alisterburt]

I don't mind about the bugfix

[alisterburt]

slice insertion can wait/come in a separate PR - no need to overload this one

[McHaillet]

awesome - really clean (but I would say that... it mirrors my own implementation 😆)

Haha, I was about to say, its mostly your implementation :p which is indeed clean!

few minor notes but happy for this to be merged when you're ready

I updated those things!

re: the 90 degree rotation I can't see anything obvious - the difference I would expect between conventions/implementations would be an inversion of rotation angle rather than a 90 degree shift, is this actually what is happening? does a zero degree projection produce an image projected along y as expected? what about 90 and x?

Double checked the projection at 0 and 90, and at 0 it is projected along the y-axis and at 90 along the x-axis. So that makes sense.

For the reconstruction, the tilt-axis is aligned with the y-axis, so to find the common line the projection from 2d to 1d needs to happen along the x-axis, i.e. the 90 addition makes sense. (the common line is perpendicular to the tilt axis)

The real-space 2d to 1d projection I wrote before in tttsa was also projecting along the x-axis after the affine transform: https://github.com/teamtomo/tttsa/blob/ecaa1bc0a92d0f17cc80c51734ebe46e045ed860/src/tttsa/projection/project_real.py#L52

So in conclusion, it is correct like this. And, I need to add 90 degrees to project along the x-axis!

[McHaillet]

For my part ready to merge, but please let me know if you have other things that should be fixed.

[alisterburt]

Glad you got to the bottom of the 90 degree offset - it makes sense and everything matches expectations here

Awesome work as always!