STEM

Scanning TEM

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
🥇 SwinIR 0.772 33.4 0.930 ✓ Certified Liang et al., ICCVW 2021
🥈 Noise2Void 0.724 31.6 0.895 ✓ Certified Krull et al., CVPR 2019
🥉 BM3D 0.635 28.5 0.820 ✓ Certified Dabov et al., IEEE TIP 2007
4 Wiener Filter 0.503 24.8 0.680 ✓ Certified Analytical baseline

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
🥇 SwinIR + gradient 0.683
0.761
31.18 dB / 0.933
0.700
27.95 dB / 0.880
0.589
23.03 dB / 0.732
✓ Certified Liang et al., ICCVW 2021
🥈 BM3D + gradient 0.649
0.673
26.19 dB / 0.837
0.661
25.89 dB / 0.829
0.614
24.26 dB / 0.778
✓ Certified Dabov et al., IEEE TIP 2007
🥉 Noise2Void + gradient 0.601
0.754
29.96 dB / 0.916
0.562
22.29 dB / 0.702
0.486
19.35 dB / 0.567
✓ Certified Krull et al., CVPR 2019
4 Wiener Filter + gradient 0.536
0.586
22.46 dB / 0.709
0.534
20.7 dB / 0.632
0.489
19.93 dB / 0.595
✓ Certified Analytical baseline

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 3 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 SwinIR + gradient 0.761 31.18 0.933
2 Noise2Void + gradient 0.754 29.96 0.916
3 BM3D + gradient 0.673 26.19 0.837
4 Wiener Filter + gradient 0.586 22.46 0.709
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_size -0.1 0.2 Å
convergence_angle -0.5 1.0 mrad
scan_distortion -0.5 1.0 %
View Details →
Dev 3 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 SwinIR + gradient 0.700 27.95 0.88
2 BM3D + gradient 0.661 25.89 0.829
3 Noise2Void + gradient 0.562 22.29 0.702
4 Wiener Filter + gradient 0.534 20.7 0.632
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_size -0.12 0.18 Å
convergence_angle -0.6 0.9 mrad
scan_distortion -0.6 0.9 %
Submit Reconstruction View Details →
Hidden 3 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 BM3D + gradient 0.614 24.26 0.778
2 SwinIR + gradient 0.589 23.03 0.732
3 Wiener Filter + gradient 0.489 19.93 0.595
4 Noise2Void + gradient 0.486 19.35 0.567
Spec Ranges (3 parameters)
Parameter Min Max Unit
probe_size -0.07 0.23 Å
convergence_angle -0.35 1.15 mrad
scan_distortion -0.35 1.15 %
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

STEM focuses the electron beam to a sub-angstrom probe and scans it across a thin specimen. The HAADF detector collects electrons scattered to large angles (>50 mrad), producing incoherent Z-contrast images where intensity scales as ~Z^1.7, enabling direct compositional interpretation at atomic resolution. Aberration correction (C3/C5 correctors) achieves sub-50 pm probe sizes. Primary degradations include scan distortion, probe instability, and radiation damage.

Principle

Scanning TEM focuses the electron beam to a fine probe (0.05-1 nm) and scans it across the specimen. Multiple detectors collect signals simultaneously: bright-field (BF), annular dark-field (ADF), and high-angle annular dark-field (HAADF). HAADF-STEM provides Z-contrast imaging where intensity scales approximately as Z^1.7, enabling direct interpretation of atomic columns by atomic number.

How to Build the System

Use an aberration-corrected STEM (probe-corrected, e.g., Thermo Fisher Titan Themis or JEOL ARM300F). Align the probe-corrector to minimize C₃ and C₅ aberrations, achieving sub-Ångström probe size. Adjust camera length for HAADF inner angle (typically 50-80 mrad for Z-contrast). Prepare atomically thin specimens by FIB or mechanical exfoliation. Use drift-corrected frame integration for high-quality atomic-resolution images.

Common Reconstruction Algorithms

  • Atom column detection and quantification (peak finding, Gaussian fitting)
  • Strain mapping via geometric phase analysis (GPA) or peak-pair analysis
  • Multi-frame averaging with rigid/non-rigid registration for noise reduction
  • HAADF simulation (frozen-phonon multislice) for quantitative comparison
  • Deep-learning STEM image denoising and super-resolution

Common Mistakes

  • Probe aberrations not fully corrected, producing probe tails and delocalization
  • Scan distortion (flyback, drift) causing apparent lattice strain artifacts
  • Sample mistilt from zone axis, reducing contrast of atomic columns
  • Amorphous surface layers (from FIB damage) obscuring atomic contrast
  • Electron channeling effects complicating quantitative HAADF interpretation

How to Avoid Mistakes

  • Tune corrector regularly using Zemlin tableau or Ronchigram analysis
  • Apply scan distortion correction using known lattice spacings as reference
  • Tilt to exact zone axis using CBED pattern or Ronchigram fine alignment
  • Use low-kV FIB final polishing or Ar-ion milling to minimize surface damage
  • Simulate HAADF images with the exact specimen thickness for quantitative analysis

Forward-Model Mismatch Cases

  • The widefield fallback applies a Gaussian PSF blur, but STEM forms images by rastering a focused electron probe (~0.1 nm) and collecting scattered electrons with annular detectors — the contrast depends on detector geometry (BF, ADF, HAADF) not optical PSF shape
  • HAADF-STEM contrast is proportional to Z^~1.7 (atomic number contrast), enabling direct chemical imaging — the widefield PSF convolution produces optical-type blur with no Z-contrast information

How to Correct the Mismatch

  • Use the STEM operator that models the electron probe profile (aberration-corrected sub-angstrom) and detector-dependent signal collection: ADF integrates scattered electrons over the annular detector range
  • For quantitative STEM, include the probe-forming aberration function, thermal diffuse scattering, and detector inner/outer angle to correctly model Z-contrast and strain mapping

Experimental Setup — Signal Chain

Experimental setup diagram for Scanning Transmission Electron Microscopy

Experimental Setup

Instrument: Nion UltraSTEM 200 / JEOL JEM-ARM200F / Thermo Fisher Titan Cubed
Accelerating Voltage Kv: 200
Convergence Semiangle Mrad: 21
Beam Current Pa: 10
Probe Size Pm: 70
Haadf Inner Angle Mrad: 80
Haadf Outer Angle Mrad: 200
Image Size: 512x512
Dwell Time Us: 20

Key References

  • Pennycook & Nellist, 'Z-Contrast STEM Imaging', Springer (2011)
  • Krivanek et al., 'Atom-by-atom structural and chemical analysis by annular dark-field electron microscopy', Nature 464, 571 (2010)

Canonical Datasets

  • NCEM Molecular Foundry STEM benchmarks
  • EMPIAR STEM datasets

Spec DAG — Forward Model Pipeline

P(e⁻) → C(probe) → D(g, η₁)

P Electron Probe (e⁻)
C Probe Formation (probe)
D ADF / BF Detector (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δd_p probe_size Probe size error (Å) 0 0.1
Δα convergence_angle Convergence semi-angle error (mrad) 0 0.5
Δs scan_distortion Scan distortion (%) 0 0.5

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.