PSF measurement, simulation, and deconvolution

[talonchandler]

• paired with this confluence page and this confluence page
• 2024-05-07 - now includes draft CLI calls psf-from-beads which estimates an averaged PSF from a bead volume, and deconvolve which uses the estimated PSF to deconvolve

TODO:

• [ ] update pre-commit

[ieivanov]

@talonchandler would you like to keep scripts/fit_psf_to_beads.py, scripts/simple_psf.py, and scripts/simulate_psf.py? They may need to have their deskew updated. These scripts call on deskew/_deskew_matrix which I think now is deprecated. Or is deskew/_get_transform_matrix the deprecated one? deskew_data uses

    matrix = _get_transform_matrix(
        raw_data.shape, ls_angle_deg, px_to_scan_ratio, keep_overhang
    )

but I think that may be a bug? I'd like to remove one of _deskew_matrix or _get_transform_matrix.

[ieivanov]

P. S. If that's a bug in the deskew, let's fix it in a separate PR that we can merge right away

[ieivanov]

P. S. If that's a bug in the deskew, let's fix it in a separate PR that we can merge right away

This is a bug in this branch only, I'll fix it

[talonchandler]

I removed scripts/fit_psf_to_beads.py, scripts/simple_psf.py, and scripts/simulate_psf.py because I don't think we'll need them. I can resurrect them if that changes.

Those are the only places that _deskew_matrix is used, so I've deleted that function, too.

With those deletions we can focus our attention on the _get_transform_matrix, which was regressing from the main branch (the GPU deskew branch changed the affine transformation matrix). I added a test to catch this problem, then fixed the problem.

I spot-tested a deskew on real data, so I now think the deskew in this branch is ready (a minor refactor from main)

@ieivanov thanks for flagging this. I just took a look, and I think these deskew functions were the only places where accidental regressions happened.

[ieivanov]

I'll test the refactored measure_psf today and then we can merge

[talonchandler]

LGTM!