``fastrotate`` optimizations? about aspire HOT 4 OPEN

Good point!
align_main is currently rather slow.
I am also using fastrotate for alignment.
How about embedding the code directly in align_main?

from aspire.

That could work. I was thinking of something more along the lines of fastrotate_fourier, where it takes a Fourier transform as input and gives a Fourier transform as output. That would save a couple of FFTs/IFFTs.

from aspire.

I looked at the code again, and I am unsure if this is possible.
For a given image, the function applies a filter along the columns (FFT+multiplication+IFFT), then along the rows, and finally, along the columns again.
I could not see how to implement this while staying in Fourier space.

from aspire.

I think I was a little bit unclear. I don't mean that we stay in Fourier space, but that we can skip the initial FFT and the final FFT, essentially creating a fastrotate_f which takes as input Fourier and outputs Fourier. This is because where it's used, align_main, only deals with the Fourier transform. Does this make sense?

from aspire.