ptychopy

Quickstart

Need to install python3 to run the GUI and ptychopy, other needed library is
in requirement.txt.(Tested OS RHEL 6.0, 7.0). This library could also be
compiled as a CUDA-C library. Inside src folder, change build.sh with your
HDF library path.

Recommend conda virtual environment, for example

conda create -n py36 python=3.6 hdf5-external-filter-plugins-lz4

1. Activate the virtual environment

source activate py36

1. To install and build the python package, set environment variables HDF5_BASE
   and CUDAHOME, which point to the installed path of the HDF5 and CUDA
   libraries. Also, set the cuda computing based on your GPU. For example the
   2080 Ti has compute capability 7.5. The GPU computing capability number can
   be found on the NVidia website

export CUDACOMPUTE=7.5
export CUDAHOME=/user/local/cuda-8.0
export HDF5_BASE=$CONDA_PREFIX

1. Then just call the following

./install.sh

1. For testing, you can use epie.py or mls.py

python epie.py

Description

This software package is implemented with ePIE, DM, MLs, reconstruction
algorithm. It supports list, spiral, cartesian, scan position file. To load a
predefined scan position file, object value files, probe value files, use the
following parameters to do the reconstruction.

API

There are two sets API for doing the reonstrunction, one set is for the whole
mode and the other set is for step mode.  The step mode is used mostly for
debug purpose, so the result for each reconstruction step could be checked. And
the whole mode would finish the whole reconstruction process and write the
result to the disk. The step mode would be slower than the whole mode, since
each step would need to transfer data back from GPU to CPU side.

The following example is based on the simulation data. For real data, the -fp
parameter has to be specified. The file format must be a bunch of HDF5 files
and the file path would be like filename_data_#06d.h5. The library will go
though all the HDF5 files and load the diffraction pattern into the library.

Whole mode API example with simulation images:

=======

Use simulation image as the example input
    ptychopy.epie(jobID="epiesimu1", beamSize=110e-9, scanDimsx=30, scanDimsy=30, stepx=50e-9, \
               stepy=50e-9, lambd=2.4796837508399954e-10, iter=3, size=512, dx_d=172e-6, z=1, simulate=1);

    ptychopy.dm(jobID="dmsimu1", beamSize=110e-9, scanDimsx=30, scanDimsy=30, stepx=50e-9, \
              stepy=50e-9, lambd=2.4796837508399954e-10, iter=3, size=512, dx_d=172e-6, z=1, simulate=1);

    ptychopy.mls(jobID="mlssimu1", beamSize=110e-9, scanDimsx=30, scanDimsy=30, stepx=50e-9, \
               stepy=50e-9, lambd=2.4796837508399954e-10, iter=3, size=512, dx_d=172e-6, z=1, simulate=0);

Use real data as the example input
    ptychopy.epie(jobID="ePIEIOTestr256", fp="/home/scan152/scan152_data_#06d.h5", \
                 fs=1, hdf5path="/entry/data/data", beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=100, size=256, dx_d=75e-6, z=1.92, dpf=51, \
                  probeModes=5)

    ptychopy.dm(jobID="DMIOTestr256", fp="/home/scan152/scan152_data_#06d.h5", \
                 fs=1, hdf5path="/entry/data/data", beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=100, size=256, dx_d=75e-6, z=1.92, dpf=51, \
                  probeModes=1)
    
    ptychopy.mls(jobID="MLSIOTestr256", fp="/home/scan152/scan152_data_#06d.h5", \
                 fs=1, hdf5path="/entry/data/data", beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=100, size=256, dx_d=75e-6, z=1.92, dpf=51, \
                  probeModes=1, delta_p=0.1, PPS=20)

Pass diffraction pattern as a 3D numpy array, diffractionNP, objectNP, probeNP

    ptychopy.epienp(jobID="ePIEIOTestr256", diffractionNP=dp,\
                 fs=1, beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=10, size=256, dx_d=75e-6, z=1.92,\
                  probeModes=2)
    
    ptychopy.dmnp(jobID="dmIOTestr256", diffractionNP=dp,\
                 fs=1, beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=10, size=256, dx_d=75e-6, z=1.92,\
                  probeModes=2)
    
    ptychopy.mlsnp(jobID="mlsIOTestr256", diffractionNP=dp,\
                 fs=1, beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=10, size=256, dx_d=75e-6, z=1.92, \
                    probeModes=2)

Step mode API example:

    epienpinit(jobID="ePIEIOTestr256", diffractionNP=dp,\
                 fs=1, beamSize=110e-6, qx=276, qy=616, scanDimsx=51, scanDimsy=51, stepx=100e-9, \
                  stepy=100e-9, lambd=1.408911284090909e-10, iter=10, size=256, dx_d=75e-6, z=1.92,\
                  probeModes=2)
    epiestep()
    episresobj()
    epiresprobe()
    epipost()

For implentation example, please check example folder.

Parameters

Reference Paper

@inproceedings{yue2021ptychopy,
  title={Ptychopy: GPU framework for ptychographic data analysis},
  author={Yue, Ke and Deng, Junjing and Jiang, Yi and Nashed, Youssef and Vine, David}
}

GitHub - kyuepublic/ptychopy: High performance ptychography reconstruction python package running on GPU

High performance ptychography reconstruction python package running on GPU - GitHub - kyuepublic/ptychopy: High performance ptychography reconstruction python package running on GPU

GitHubkyuepublic