fold_slice PrScO3

(Top)

Part of a series on Phase retrieval

Setup

0. Install MATLAB
   Required Add-ons for fold_slice:

   • Signal Processing
   • Parallel Computing
   • Curve Fitting
   • Image Processing
   • Machine Learning and Statistics
   • Optimization

1. get fold_slice code

   git clone https://github.com/yijiang1/fold_slice.git

Data

PrScO3 sample data link

Save sample_data_PrScO3.mat file (and experimental_conditions.txt for more info/post-processing)
(Recommended: save to separate project_folder/data/ directory)

Running example script

1. go to fold_slice/ptycho/ directory

2. copy all examples/PSO_science/xxxx.m files to ptycho/ directory

3. First open prepare_data_PSO_science.m

   1. change data_dir to your data directory (line 9)
   2. run all blocks

4. Open ptycho_electron_PSO_science.m

   1. change base_path to your project directory (line 6)
   2. run all blocks
      • Most errors (shown as #### PONG! ####) will probably occur here. See troubleshooting.

   An explanation of the code structure can be found here or the fold_slice github

5. Check results

   • results are in base_path/1/ folder
   • The xxxx.hdf5 file is for PtychoShelves data management
   • roi0_Ndp128/ folder (region of interest 0, 128x128 detector array) has reconstructions
     • Folder names have information on recon method
     • MLs_L1_p8_g64_pc50_noModel_vp1_Ns21_dz10_reg1/
       • MLs: Maximum Likelihood
       • L1: error metric = L1 (왜 L2 안쓰는지는 모르겠음)
       • p8: probe modes = 8
       • g64: batch size = 64
       • pc50: probe position correction starts at iteration 50
       • noModel: no initial object guess
       • vp1: variable probe = True
       • Ns21: num slices = 21
       • dz10: slice separation = 10 Angstroms
     • MLs_L1_p8_g32_Ndp256_pc50_noModel_vp1_Ns21_dz10_reg1/
       • Ndp256: oversample 128x128 detector array to 256x256 (artificially “increases” resolution, but with more artefacting)

Troubleshooting

• function xxxx not found type errors:
  • Always run the code in ptycho directory
  • Search function name - sometimes they are Toolbox functions that are not installed
• CUDA errors
  • Probably version problems with Windows C++/CUDA compilers.
  • Never succeeded running on windows…
  • Workaround solution: Windows subsystem for Linux: Install MATLAB inside WSL

  A note on Windows and CUDA

  PtychoShelves’ reconsturction engines are usually built as C++ subroutines, probably for faster computation. They call for Matlab mex to compile these at execution time.

  The problem:
  MinGW, the conventional C++ compiler for Windows, does not support CUDA. Searching online, it seems that you need Visual Studio and its official compilers by Nvidia. I gave up on this approach - please let me know if there is another way to compile for CUDA on Windows.

• on CPU
  • 지못미

Notes

PtychoShelves

PtychoShelves is a Matlab framework for ptychography - it is a pipeline that allows the use of various reconstruction algorithms.

PtychoShelves uses a modular approach, you can choose any process for file i/o, data preprocessing, reconstruction algorithm, etc. Popular algorithms for ptychography including Maximum Likelihood and Difference Map are already implemented. You can also make your own processes as well.

This pipeline is realized through the p struct, where you can assign each module and their (hyper)parameters for the pipeline. The individual subroutines can be implemented/accessed via their own structs, for example the det struct for detector settings and the eng struct for the reconstruction engine settings.

I’m not familiar with this yet; need to get used to changing p attributes and implementing custom engines.

fold_slice

Chen, Z. et al. (2021) uses a version of fold_slice, a package built on top of PtychoShelves made for multislice ptychography.

The major differences being - Data and reconstruction I/O conventions changed for electron ptychography (original Ptychoshelves is mainly based on X-ray ptychography) - A modified least-squares maximum likelihood multislice reconstruction engine implemented in GPU_MS

A rough structure of the code can be found here, and an explanation of the reconstruction here.

See also

Published papers

• Multislice EM, PrScO3
  • Chen, Z. et al. Electron ptychography achieves atomic-resolution limits set by lattice vibrations. Science 372, 826–831 (2021).
• PtychoShelves Matlab framework
  • Wakonig, K. et al. PtychoShelves, a versatile high-level framework for high-performance analysis of ptychographic data. J Appl Cryst 53, 574–586 (2020).
  • (unmodified) PtychoShelves package download link
• Maximum likelihood
  • Thibault, P. & Guizar-Sicairos, M. Maximum-likelihood refinement for coherent diffractive imaging. New J. Phys. 14, 063004 (2012).
    
  • Odstrčil, M., Menzel, A. & Guizar-Sicairos, M. Iterative least-squares solver for generalized maximum-likelihood ptychography. Opt. Express, OE 26, 3108–3123 (2018).
• Mixed state probe reconstruction
  • Thibault, P. & Menzel, A. Reconstructing state mixtures from diffraction measurements. Nature 494, 68–71 (2013).
• Multislice
  • E. H. R. Tsai, I. Usov, A. Diaz, A. Menzel, and M. Guizar-Sicairos, X-ray ptychography with extended depth of field, Opt. Express 24, 29089-29108 (2016).

Other helpful sources

• fold_slice github
  • Yi Jiang’s modified PtychoShelves → multislice ptychography
  • There is a link to ptychography study group videos/materials in README