FPM Workflow Overview¶

What is Fourier ptychography?¶

Fourier Ptychographic Microscopy (FPM) uses an LED array to illuminate a sample from different angles. Each LED illuminates the sample with a plane wave at a specific angle, producing a low-resolution image on a camera. By combining many such images from different angles, FPM computationally synthesizes a high-resolution, wide-field image — recovering both amplitude and phase — without mechanically scanning the sample.

Data flow¶

HDF5 File
    |
    v
ExperimentalData   (loads intensity images, LED positions, geometry)
    |
    v
IlluminationCalibration   (optional: refine LED positions/angles)
    |
    v
Reconstruction     (holds mutable state: object, probe/pupil)
    |
    v
Engine.reconstruct()   (iterative phase retrieval in Fourier domain)
    |
    v
Monitor            (real-time visualization)

FPM vs CPM¶

	CPM	FPM
Illumination	Coherent focused probe scanned across sample	LED array: angular illumination scanning
Scanning	Sample plane (real space)	Fourier plane (pupil function)
Output	Complex object + probe in real space	Complex object + pupil function
Resolution	Limited by probe overlap	Limited by LED array NA

Step-by-step FPM workflow¶

1. Initialize with `easyInitialize`¶

import PtyLab
from PtyLab import Engines

data, recon, params, monitor, engine, calib = PtyLab.easyInitialize(
    "fpm_data.hdf5",
    engine=Engines.mqNewton,
    operationMode="FPM",
)

easyInitialize returns a 6-tuple for FPM — note the additional IlluminationCalibration object:

Object	Description
`experimentalData`	FPM diffraction data and geometry
`reconstruction`	Mutable state: object array, pupil
`params`	Shared configuration
`monitor`	Real-time visualization
`engine`	Reconstruction algorithm
`calib`	LED calibration helper (`IlluminationCalibration`)

2. Calibrate LED positions (optional but recommended)¶

Real LED arrays often have small positioning errors. The IlluminationCalibration object fits a model to correct these:

calib.fit_mode = "Translation"
calib.runCalibration()

See LED Calibration for details.

3. Reconstruct¶

recon.initializeObjectProbe()
engine.numIterations = 50
for loop, posLoop in engine.reconstruct():
    pass

recon.saveResults("fpm_result.hdf5")

FPM required data fields¶

Field	Type	Unit	Description
`ptychogram`	`float32 (N, Nd, Nd)`	counts	Stack of N intensity images
`wavelength`	`float64`	m	Illumination wavelength
`encoder`	`float64 (N, 2)`	m	LED (x, y) positions
`dxd`	`float64`	m	Camera pixel size (in the object plane after magnification)
`zled`	`float64`	m	LED-array to sample distance
`magnification`	`float64`	—	Microscope magnification

FPM optional data fields¶

Field	Type	Description
`NA`	`float64`	Numerical aperture of the objective

Recommended engines for FPM¶

Engine	Notes
`mqNewton`	Recommended — fast quasi-Newton updates
`mPIE`	Good all-purpose choice with momentum
`ePIE`	Simplest; use for debugging