Live Iterative Ptychography with projection-based algorithms

animation_compr.mp4

Accompanying code repository for our work Live Iterative Ptychography with projection-based algorithms. Developed by Simon Welker and Tal Peer, based in part on the bachelor thesis of Johannes Kolhoff.

We kindly ask you to cite our work in your publication when using any of our research or code:

@article{welker2023livePty,
  title={Live Iterative Ptychography with projection-based algorithms},
  author={Welker, Simon and Peer, Tal and Chapman, Henry N. and Gerkmann, Timo},
  journal={arXiv preprint arXiv:2309.08639},
  date={2023-09-14},
  eprint={2309.08639},
  eprinttype={arxiv},
  pubstate={preprint},
}

Setup

We recommend using Python 3.9 or newer. Create a new virtual environment and run

pip install -r requirements.txt

Data preparation

If you wish to use our simulated dataset based on DIV2K, you first need to download the DIV2K validation set from the following URL, extract it, and then run the following ImageMagick command inside your copy of the DIV2K/DIV2K_valid_HR/ folder:

for f in *.png; do convert "$f" -resize 512x512^ -gravity Center -crop 512x512+0+0 +repage "512crop/${f}"; done

which generates a folder of all images resized and center-cropped to 512x512 (as we used it). Then run:

python prepare_data.py <the_512crop_subfolder_you_just_generated>/

which generates all 90 test objects and probes and stores them in the file data/objects_probes.h5 of this repository.

Repository structure

• Library code is inside src/
• Scripts to run reconstructions are:
  • run_recon.py (live reconstruction)
  • run_recon_classical.py (non-live reconstruction)
  • run_central_recons.py (for pre-reconstruction of the central region).
• Results will be stored as HDF5 files under results/
  • view_recon.py can be used for visualization of a single result file (e.g. results/run1/0.h5). We recommend using the options --zero-phase-ramp --apply-gamma for best viewing experience.
• Data is stored inside data/
  • This data is not provided with the repository and must be generated, via prepare_data.py (for simulated test objects and probes) and run_central_recons.py (for pre-reconstructed central regions).