Segmented and Other Geometry Detector Ptychography

[juliedactyl]

This PR adds support for performing ptychography with arbitrary detectors, including segmented annular detectors and lower number pixel arrays.

[gvarnavi]

Ooops, tried to extend the virtual detector ptycho functionality to all classes during the flight and seem to have messed up my git commands to push to someone else's PR 😬.

Closing this one and opening a new one. Apologies -- commits should still be credited correctly.

[juliedactyl]

This is now ready for review.

Pasting George's comment from PR #684:

Taking a step back, this essentially adds the following snippet to the ptycho Fourier-projection step:

      if virtual_detector_masks is not None:
          masked_values = xp.sum(
              fourier_overlap[:, None, :, :] * virtual_detector_masks[None, :, :, :],
              axis=(-1, -2),
          ).transpose()
          fourier_overlap = xp.zeros_like(fourier_overlap)
          for mask, value in zip(virtual_detector_masks, masked_values):
              fourier_overlap[:, mask] = value[:, None] / xp.sum(mask)

which will bin the forward-model exit-waves using arbitrary (corner-centered) virtual_detector_masks. If the dataset is similarly binned (e.g. by preprocessing a segmented detector dataset), then it provides a flexible way of reconstructing specific forward models.

Attaching a notebook to help review, which illustrates the functionality for an annular segmented detector:

virtual-detector-ptycho-PR684.zip

[smribet]

Hi @juliedactyl and @gvarnavi, thanks for the PR and test notebook! Look forward to trying this out!