Ptychography

Ptychographic Imaging

Join the Competition Contribute to Benchmark

Standard reconstruction benchmark — forward model perfectly known, no calibration needed. Score = 0.5 × clip((PSNR−15)/30, 0, 1) + 0.5 × SSIM

# Method Score PSNR (dB) SSIM Source
🥇 AutoPhaseNN 0.784 34.0 0.935 ✓ Certified Chan et al., 2024
🥈 PtychoNN 0.747 32.5 0.910 ✓ Certified Cherukara et al., 2020
🥉 sDR 0.635 28.5 0.820 ✓ Certified Wen et al., J. Opt. 2019
4 ePIE 0.522 25.0 0.710 ✓ Certified Maiden & Rodenburg, 2009

Dataset: PWM Benchmark (4 algorithms)

Blind Reconstruction Challenge — forward model has unknown mismatch, must calibrate from data. Score = 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

# Method Overall Score Public
PSNR / SSIM
 Dev
PSNR / SSIM
 Hidden
PSNR / SSIM
 Trust Source
🥇 AutoPhaseNN + gradient 0.666
0.770
31.95 dB / 0.942
0.637
24.65 dB / 0.791
0.590
23.12 dB / 0.736
✓ Certified Chan et al., Commun. Phys. 2024
🥈 sDR + gradient 0.661
0.700
26.92 dB / 0.856
0.666
26.35 dB / 0.842
0.616
24.67 dB / 0.792
✓ Certified Wen et al., J. Opt. 2019
🥉 PtychoNN + gradient 0.643
0.745
29.87 dB / 0.915
0.648
25.43 dB / 0.815
0.536
21.06 dB / 0.648
✓ Certified Cherukara et al., Appl. Phys. Lett. 2020
4 ePIE + gradient 0.586
0.590
22.61 dB / 0.715
0.606
23.54 dB / 0.752
0.562
22.36 dB / 0.705
✓ Certified Maiden & Rodenburg, Ultramicroscopy 2009

Complete score requires all 3 tiers (Public + Dev + Hidden).

Join the competition →
Scoring: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖) PSNR 40% · SSIM 40% · Consistency 20%
Public 5 scenes

Full-access development tier with all data visible.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

How to use: Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

What to submit: Reconstructed signals (x_hat) and corrected spec as HDF5.

Public Leaderboard
# Method Score PSNR SSIM
1 AutoPhaseNN + gradient 0.770 31.95 0.942
2 PtychoNN + gradient 0.745 29.87 0.915
3 sDR + gradient 0.700 26.92 0.856
4 ePIE + gradient 0.590 22.61 0.715
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_error -5.0 10.0 %
position_error -10.0 20.0 nm
partial_coherence -5.0 10.0 nm
View Details →
Dev 5 scenes

Blind evaluation tier — no ground truth available.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), and spec ranges only.

How to use: Apply your pipeline from the Public tier. Use consistency as self-check.

What to submit: Reconstructed signals and corrected spec. Scored server-side.

Dev Leaderboard
# Method Score PSNR SSIM
1 sDR + gradient 0.666 26.35 0.842
2 PtychoNN + gradient 0.648 25.43 0.815
3 AutoPhaseNN + gradient 0.637 24.65 0.791
4 ePIE + gradient 0.606 23.54 0.752
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_error -6.0 9.0 %
position_error -12.0 18.0 nm
partial_coherence -6.0 9.0 nm
Submit Reconstruction View Details →
Hidden 5 scenes

Fully blind server-side evaluation — no data download.

What you get & how to use

What you get: No data downloadable. Algorithm runs server-side on hidden measurements.

How to use: Package algorithm as Docker container / Python script. Submit via link.

What to submit: Containerized algorithm accepting y + H, outputting x_hat + corrected spec.

Hidden Leaderboard
# Method Score PSNR SSIM
1 sDR + gradient 0.616 24.67 0.792
2 AutoPhaseNN + gradient 0.590 23.12 0.736
3 ePIE + gradient 0.562 22.36 0.705
4 PtychoNN + gradient 0.536 21.06 0.648
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_error -3.5 11.5 %
position_error -7.0 23.0 nm
partial_coherence -3.5 11.5 nm
Submit Algorithm View Details →

Blind Reconstruction Challenge

Challenge

Given measurements with unknown mismatch and spec ranges (not exact params), reconstruct the original signal. A method must be evaluated on all three tiers for a complete score. Scored on a composite metric: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖).

Input

Measurements y, ideal forward model H, spec ranges

Output

Reconstructed signal x̂

About the Imaging Modality

Ptychography is a scanning coherent diffractive imaging technique where a coherent beam (X-ray or electron) illuminates overlapping regions of the sample and far-field diffraction patterns are recorded at each scan position. The overlap between adjacent probe positions provides redundancy that enables simultaneous recovery of the complex-valued object transmission function and the illumination probe via iterative algorithms (ePIE, difference map). The forward model at each position is I_j = |F{P(r-r_j) * O(r)}|^2 where P is the probe and O is the object. Achievable resolution is limited by the detector NA, not the optics, reaching sub-10 nm for X-rays.

Principle

Ptychography is a scanning coherent diffractive imaging technique where a coherent beam (visible, X-ray, or electron) illuminates overlapping regions of the sample. At each scan position, a far-field diffraction pattern is recorded. The redundancy from overlapping illumination positions constrains the phase-retrieval problem, enabling simultaneous recovery of both the complex sample transmittance and the illumination probe function.

How to Build the System

For X-ray ptychography at a synchrotron: focus the beam to a defined spot (0.1-1 μm) using a Fresnel zone plate or KB mirrors. Mount the sample on a precision piezo scanning stage. Place a photon-counting area detector (Eiger, Pilatus) in the far field (1-5 m downstream). Scan positions should overlap by 60-70 %. For visible-light or electron ptychography, adapt the geometry but maintain the overlap requirement.

Common Reconstruction Algorithms

  • ePIE (extended Ptychographic Iterative Engine)
  • Difference Map algorithm
  • Maximum Likelihood refinement (MLR)
  • PtychoShelves (modular framework for ptychographic reconstruction)
  • Deep-learning ptychography (PtychoNN, learned phase retrieval)

Common Mistakes

  • Insufficient overlap between adjacent scan positions (need ≥60 %)
  • Position errors in the scanning stage causing reconstruction artifacts
  • Partial coherence effects not modeled, degrading recovered phase
  • Vibration or drift during the scan corrupting the diffraction data
  • Detector saturation at the central beam stop region

How to Avoid Mistakes

  • Maintain ≥65 % overlap; include position correction in the reconstruction algorithm
  • Use position refinement (annealing) as part of the ptychographic reconstruction
  • Include mixed-state (multi-mode) probe to model partial coherence
  • Use interferometric position feedback and short dwell times per point
  • Use a semi-transparent beam stop or high-dynamic-range detector modes

Forward-Model Mismatch Cases

  • The widefield fallback produces a single (64,64) image, but ptychography acquires diffraction patterns at multiple overlapping scan positions — output shape (n_positions, det_x, det_y) is a set of far-field intensity measurements
  • Ptychography is fundamentally nonlinear (y_j = |F{P * O_j}|^2, intensity of Fourier transform of probe times object) — the widefield linear blur cannot model coherent wave propagation, diffraction, or phase retrieval

How to Correct the Mismatch

  • Use the ptychography operator that generates one far-field diffraction pattern per probe position, with overlapping illumination enabling redundant phase information for robust reconstruction
  • Reconstruct using PIE (Ptychographic Iterative Engine), ePIE, or gradient-descent methods that alternate between real-space (overlap constraint) and Fourier-space (modulus constraint) using the coherent forward model

Experimental Setup — Signal Chain

Experimental setup diagram for Ptychographic Imaging

Experimental Setup

Instrument: Diamond Light Source I13 / APS 2-ID / ESRF ID16A
Photon Energy Kev: 12.4
Wavelength Nm: 0.1
Detector: Eiger 500K (512x512 px, 75 um pitch)
Probe Size Um: 1.0
Step Size Nm: 200
Overlap Ratio: 0.7
Propagation Distance M: 2.1
Achieved Resolution Nm: 10
Reconstruction: ePIE / difference map / Adam-based optimization

Key References

  • Rodenburg & Faulkner, 'A phase retrieval algorithm for shifting illumination (ePIE)', Appl. Phys. Lett. 85, 4795-4797 (2004)
  • Thibault et al., 'High-resolution scanning X-ray diffraction microscopy', Science 321, 379-382 (2008)

Canonical Datasets

  • PtychoNN benchmark datasets (Cherukara et al.)
  • Diamond I13 ptychography test data

Spec DAG — Forward Model Pipeline

P(probe) → D(g, η₁)

P Probe Illumination + Propagation (probe)
D Diffraction Detector (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
ΔP probe_error Probe function error (%) 0 5.0
Δr position_error Scan position error (nm) 0 10.0
σ_c partial_coherence Partial coherence width (nm) 0 5.0

Credits System

40%
Platform Profit Pool
Revenue allocated to benchmark rewards
30%
Winner Share
Top algorithm receives from pool
$100
Min Withdrawal
Minimum payout threshold
Spec Primitives Reference (11 primitives)
P Propagation

Free-space or medium propagation kernel (Fresnel, Rayleigh-Sommerfeld).

M Mask / Modulation

Spatial or spatio-temporal amplitude modulation (coded aperture, SLM pattern).

Π Projection

Geometric projection operator (Radon transform, fan-beam, cone-beam).

F Fourier Sampling

Sampling in the Fourier / k-space domain (MRI, ptychography).

C Convolution

Shift-invariant convolution with a point-spread function (PSF).

Σ Summation / Integration

Summation along a physical dimension (spectral, temporal, angular).

D Detector

Sensor readout with gain g and noise model η (Gaussian, Poisson, mixed).

S Structured Illumination

Patterned illumination (block, Hadamard, random) applied to the scene.

W Wavelength Dispersion

Spectral dispersion element (prism, grating) with shift α and aperture a.

R Rotation / Motion

Sample or gantry rotation (CT, electron tomography).

Λ Wavelength Selection

Spectral filter or monochromator selecting a wavelength band.