Skip to content

Full radios processing pipeline: step 1

Pierre Paleo requested to merge fullfield_mixed into master

About

This MR implements a full-field pipeline dedicated to processing entire radios ("Solution 1" of #223 (closed)). The Cuda backend will be done in another MR.

Close #223 (closed)

To do

  • FullRadiosPipeline
  • FullRadiosLocalReconstruction: radios processing
  • SinosStackPipeline
  • FullRadiosLocalReconstruction: sinos processing
  • Combine the two using in-memory data
  • Combine the two using on-disk data
  • Complete docstrings
  • End-to-end reconstruction test (use cases: rotation, CTF)

Notes

  • This MR also adds the support for FFTW in Paganin phase retrieval, as it is a noticeable bottleneck when the pipeline is executed on CPU.
  • Steps ccd_correction and unsharp_mask are fearfully slow on CPU. But they are easy to parallelize using multiprocessing.pool.ThreadPool. It might be interesting to add this in a future MR.
Edited by Pierre Paleo

Merge request reports