02 – CPM Example Data¶

This tutorial demonstrates a complete CPM reconstruction workflow using a real experimental dataset loaded from an HDF5 file. It covers loading data, initializing the reconstruction, configuring parameters, and running an iterative engine.

What you'll learn:

• How to load experimental CPM data from an HDF5 file
• How to configure Params and choose a reconstruction engine
• How to iterate and visualize the reconstruction in real time

!!! note "Dataset" The dataset (simu.hdf5) is generated automatically into the example_data/ folder at the project root the first time you run the notebook. If the file already exists it is not re-generated.

!!! tip "Prerequisites" Complete 01 – CPM Simulation before this tutorial.

In [1]:

import matplotlib
import matplotlib.pyplot as plt
import numpy as np

# import the PtyLab module
import PtyLab

import matplotlib import matplotlib.pyplot as plt import numpy as np # import the PtyLab module import PtyLab

In [ ]:

from PtyLab.io import getExampleDataFolder, generate_simu_hdf5

fileName = "simu.hdf5"
filePath = getExampleDataFolder() / fileName

if not filePath.exists():
    print("Generating simu.hdf5...")
    generate_simu_hdf5(filePath)
    print("Done.")
else:
    print(f"{fileName} found, skipping generation.")

from PtyLab.io import getExampleDataFolder, generate_simu_hdf5 fileName = "simu.hdf5" filePath = getExampleDataFolder() / fileName if not filePath.exists(): print("Generating simu.hdf5...") generate_simu_hdf5(filePath) print("Done.") else: print(f"{fileName} found, skipping generation.")

simu.hdf5 found, skipping generation.

There's an easyInitializa method that you can call to initialize all the classes needed for a reconstruction. To do that you can uncomment the code below.

But we recommend you to initialize each class one by one and follow the tutorial below.

In [3]:

## optional
exampleData, reconstruction, params, monitor, ePIE_engine = PtyLab.easyInitialize(
    filePath
)

## optional exampleData, reconstruction, params, monitor, ePIE_engine = PtyLab.easyInitialize( filePath )

INFO:GPU:cupy and CUDA available, switching to GPU
INFO:Reconstruction:Copying attribute wavelength
INFO:Reconstruction:Copying attribute dxd
INFO:Reconstruction:Copying attribute theta
INFO:Reconstruction:Copying attribute spectralDensity
INFO:Reconstruction:Copying attribute entrancePupilDiameter
INFO:Reconstruction:Initial object set to ones
INFO:Reconstruction:Initial probe set to circ
INFO:ePIE:Sucesfully created ePIE ePIE_engine
INFO:ePIE:Wavelength attribute: 6.327999813038332e-07

Found encoder with shape (100, 2)

ExperimentalData class¶

The data must be stored as a ".hdf5" file, which enables structured file storage. For ptychographic data, experimental parameters such as wavelength or pixel size can be convieniently stored together with illumination/ecoder positions and the actual raw images/diffraction data in a single file.

A minimal list of fields required for ptyLab to work are:

• ptychogram - 3D image/diffraction data stack
• wavelength - illumination lambda
• encoder - encoder positions / illumination angles
• dxd - detector pixel size
• zo - sample-detector distance

Also we have optional fields because they will either be computed later from the "required_fields" or are required for FPM, but not CPM (or vice-versa). If not provided by the user they will be set as None.

• dxp - can be provided by the user (otherwise will be computed using dxp=dxd/magnification for FPM)
• No - number of upsampled pixels
• Nd - probe/pupil plane size, will be set to Ptychogram size by default
• entrancePupilDiameter - In CPM, put the estimate of the diameter of the beam in this variable. (different usage in FPM, check the tutorial for FPM)
• spectralDensity - a list wavelength for multi wavelength reconsruction
• theta - tilt angle for reflection mode CPM
• magnification - magnification, used for FPM computations of dxp

The ".hdf5" file must contain a field called "ptychogram" containing the experimental raw images as a 3D array of shape [numFrames,X,Y], where numFrames is the number of images corresponding to each illumination vector in the "encoder" and X-Y are the 2D image dimensions.

The ".hdf5" file must have a field called "encoder" containing the translation stage positions in units of meters (for CP) or the illumination angles in units of rad (for FP). The field "encoder" has a 2D shape [numFrames,2], where numFrames is the number of positions.

We start off our demonstration by creating the ExperimentalData() class which is used to load the .hdf5 file. In this example the variable "exampleData" will contain our class.

In [4]:

# initialize the ExperimentalData class
exampleData = PtyLab.ExperimentalData(filePath)

# initialize the ExperimentalData class exampleData = PtyLab.ExperimentalData(filePath)

Found encoder with shape (100, 2)

In [5]:

# now, all our experimental data is loaded into exampleData, and we can check our diffraction patterns (or ptychograms).
exampleData.showPtychogram()  # close the figure to proceed

# now, all our experimental data is loaded into exampleData, and we can check our diffraction patterns (or ptychograms). exampleData.showPtychogram() # close the figure to proceed

Min max ptychogram: 2.462452171423518e-14, 32774.45703125
Min max ptychogram: 0.0, 4.5155487060546875
0.0 4.5155487

VBox(children=(IntSlider(value=0, description='Frame', max=99), Output()))

Reconstruction class¶

The Reconstruction class creates an object which will be mutable during the reconstruction. We first initialize the reconstruction class, thereby coping necessary attributes from the exampleData to the reconstruction object.

In [6]:

## initialize the Reconstruction class
reconstruction = PtyLab.Reconstruction(exampleData, params)

## initialize the Reconstruction class reconstruction = PtyLab.Reconstruction(exampleData, params)

INFO:Reconstruction:Copying attribute wavelength
INFO:Reconstruction:Copying attribute dxd
INFO:Reconstruction:Copying attribute theta
INFO:Reconstruction:Copying attribute spectralDensity
INFO:Reconstruction:Copying attribute entrancePupilDiameter

Initialize the probe and object (nlambda,nosm,npsm,nslice,Ny,Nx), default values for the first four axes are 1.

In [7]:

# now create an object to hold everything we're eventually interested in
reconstruction.npsm = 1  # Number of probe modes to reconstruct
reconstruction.nosm = 1  # Number of object modes to reconstruct
reconstruction.nlambda = 1  # len(exampleData.spectralDensity) # Number of wavelength
reconstruction.nslice = 1  # Number of object slice

# set initial guesses
reconstruction.initialProbe = "circ"
reconstruction.initialObject = "ones"
# initialize probe and object and related params
reconstruction.initializeObjectProbe()

# customize initial probe quadratic phase if wish
reconstruction.probe = reconstruction.probe * np.exp(
    1.0j
    * 2
    * np.pi
    / reconstruction.wavelength
    * (reconstruction.Xp**2 + reconstruction.Yp**2)
    / (2 * 6e-3)
)

# now create an object to hold everything we're eventually interested in reconstruction.npsm = 1 # Number of probe modes to reconstruct reconstruction.nosm = 1 # Number of object modes to reconstruct reconstruction.nlambda = 1 # len(exampleData.spectralDensity) # Number of wavelength reconstruction.nslice = 1 # Number of object slice # set initial guesses reconstruction.initialProbe = "circ" reconstruction.initialObject = "ones" # initialize probe and object and related params reconstruction.initializeObjectProbe() # customize initial probe quadratic phase if wish reconstruction.probe = reconstruction.probe * np.exp( 1.0j * 2 * np.pi / reconstruction.wavelength * (reconstruction.Xp**2 + reconstruction.Yp**2) / (2 * 6e-3) )

INFO:Reconstruction:Initial object set to ones
INFO:Reconstruction:Initial probe set to circ

Monitor class¶

This class will create a monitor to visualize the reconstruction.

In [8]:

## Initialise the monitor class
monitor = PtyLab.Monitor()

## Initialise the monitor class monitor = PtyLab.Monitor()

In [9]:

## Set monitor properties
monitor.verboseLevel = "low"  # low (default): plot only one figure (object/probe/error), high: add a second figure showing measured and estimated diffraction patterns
monitor.figureUpdateFrequency = 1  # the frequency of the plots
monitor.objectPlot = "complex"  # options: complex, abs, angle
monitor.objectZoom = 1.5  # control object plot FoV
monitor.probeZoom = 0.5  # control probe plot FoV
monitor.objectPlotContrast = 0.9  # control the contrast of object plot, normalized intensity is from [0,1], objectPlotContrast sets the maximum value for color axis
monitor.probePlotContrast = 0.9  # control the contrast of probe plot

## Set monitor properties monitor.verboseLevel = "low" # low (default): plot only one figure (object/probe/error), high: add a second figure showing measured and estimated diffraction patterns monitor.figureUpdateFrequency = 1 # the frequency of the plots monitor.objectPlot = "complex" # options: complex, abs, angle monitor.objectZoom = 1.5 # control object plot FoV monitor.probeZoom = 0.5 # control probe plot FoV monitor.objectPlotContrast = 0.9 # control the contrast of object plot, normalized intensity is from [0,1], objectPlotContrast sets the maximum value for color axis monitor.probePlotContrast = 0.9 # control the contrast of probe plot

/mnt/home/shantanu/projects/PtyLab.py/PtyLab/Monitor/Monitor.py:178: UserWarning: For faster update of the reconstruction plot, set `monitor.figureUpdateFrequency = 5` or higher.
  warnings.warn(

Params class¶

This class holds the parameters that determine how the reconstruction is performed. For instance to determine whether the reconstruction is carried out on CPU or GPU, or what propagator we want to use.

Various switches and parameters of different kinds of regularizations are also specified before a reconstruction is carried out.

In [10]:

## Initialise Params class
params = PtyLab.Params()

## Initialise Params class params = PtyLab.Params()

INFO:GPU:cupy and CUDA available, switching to GPU

In [11]:

## main parameters
params.positionOrder = "random"  # 'sequential' or 'random'
params.propagator = (
    "Fresnel"  # Fraunhofer Fresnel ASP scaledASP polychromeASP scaledPolychromeASP
)

## how do we want to reconstruct?
params.gpuSwitch = True  # Turn on if gpu is available, with cupy imported.
params.probePowerCorrectionSwitch = (
    True  # fix the power of the probe to the maximum of the ptychogram
)
params.comStabilizationSwitch = (
    True  # fix the center-of-mass of the probe in the center of the probe frame
)
params.orthogonalizationSwitch = (
    False  # turn on when performing mixed states reconstructions
)
params.orthogonalizationFrequency = 10  # the frequency of performing orthogonalization
params.fftshiftSwitch = False
params.intensityConstraint = "standard"  # standard fluctuation exponential poission
params.absorbingProbeBoundary = False
params.objectContrastSwitch = False
params.absObjectSwitch = False
params.backgroundModeSwitch = False
params.couplingSwitch = True
params.couplingAleph = 1
params.positionCorrectionSwitch = False

## main parameters params.positionOrder = "random" # 'sequential' or 'random' params.propagator = ( "Fresnel" # Fraunhofer Fresnel ASP scaledASP polychromeASP scaledPolychromeASP ) ## how do we want to reconstruct? params.gpuSwitch = True # Turn on if gpu is available, with cupy imported. params.probePowerCorrectionSwitch = ( True # fix the power of the probe to the maximum of the ptychogram ) params.comStabilizationSwitch = ( True # fix the center-of-mass of the probe in the center of the probe frame ) params.orthogonalizationSwitch = ( False # turn on when performing mixed states reconstructions ) params.orthogonalizationFrequency = 10 # the frequency of performing orthogonalization params.fftshiftSwitch = False params.intensityConstraint = "standard" # standard fluctuation exponential poission params.absorbingProbeBoundary = False params.objectContrastSwitch = False params.absObjectSwitch = False params.backgroundModeSwitch = False params.couplingSwitch = True params.couplingAleph = 1 params.positionCorrectionSwitch = False

Engine class¶

A specific engine can be imported (e.g. ePIE, mPIE etc.) to optimize our initial estimates for the object/probe.

In [12]:

## Initialize the engine class by passing the exampleData, reconstruction, params, and monitor objects to the chosen engine
ePIE = PtyLab.Engines.ePIE(reconstruction, exampleData, params, monitor)

## Initialize the engine class by passing the exampleData, reconstruction, params, and monitor objects to the chosen engine ePIE = PtyLab.Engines.ePIE(reconstruction, exampleData, params, monitor)

INFO:ePIE:Sucesfully created ePIE ePIE_engine
INFO:ePIE:Wavelength attribute: 6.327999813038332e-07

In [13]:

ePIE.numIterations = 10
ePIE.betaObject = 0.25
ePIE.betaProbe = 0.25
ePIE.reconstruct()

ePIE.numIterations = 10 ePIE.betaObject = 0.25 ePIE.betaProbe = 0.25 ePIE.reconstruct()

Out[13]:

<generator object ePIE.reconstruct at 0x7d343c778040>

In [14]:

## switch engine if you need
# switch to mPIE
mPIE = PtyLab.Engines.mPIE(reconstruction, exampleData, params, monitor)
mPIE.numIterations = 100
mPIE.betaProbe = 0.25
mPIE.betaObject = 0.25
mPIE.reconstruct()

## switch engine if you need # switch to mPIE mPIE = PtyLab.Engines.mPIE(reconstruction, exampleData, params, monitor) mPIE.numIterations = 100 mPIE.betaProbe = 0.25 mPIE.betaObject = 0.25 mPIE.reconstruct()

mPIE: 100%|██████████| 100/100 [00:22<00:00,  4.49it/s]

INFO:mPIE:switch to cpu

In [15]:

## now save the data
reconstruction.saveResults("reconstruction.hdf5")

## now save the data reconstruction.saveResults("reconstruction.hdf5")

The reconstruction results (all) have been saved