SEM

Scanning Electron Microscopy

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 0.713
0.763
31.55 dB / 0.938
0.716
29.08 dB / 0.902
0.661
26.38 dB / 0.842
✓ Certified Liang et al., ICCVW 2021
🥈 Noise2Void + gradient 0.567
0.735
29.74 dB / 0.913
0.525
20.46 dB / 0.621
0.441
17.75 dB / 0.488
✓ Certified Krull et al., CVPR 2019
🥉 BM3D + gradient 0.565
0.669
25.73 dB / 0.824
0.564
22.23 dB / 0.700
0.463
18.42 dB / 0.521
✓ Certified Dabov et al., IEEE TIP 2007
4 Wiener Filter + gradient 0.557
0.618
23.39 dB / 0.746
0.569
22.26 dB / 0.701
0.484
19.27 dB / 0.563
✓ 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 0.763 31.55 0.938
2 Noise2Void + gradient 0.735 29.74 0.913
3 BM3D + gradient 0.669 25.73 0.824
4 Wiener Filter + gradient 0.618 23.39 0.746
Spec Ranges (3 parameters)
Parameter Min Max Unit
beam_energy -0.1 0.2 keV
stigmatism -5.0 10.0 nm
working_distance -0.1 0.2 mm
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 0.716 29.08 0.902
2 Wiener Filter + gradient 0.569 22.26 0.701
3 BM3D + gradient 0.564 22.23 0.7
4 Noise2Void + gradient 0.525 20.46 0.621
Spec Ranges (3 parameters)
Parameter Min Max Unit
beam_energy -0.12 0.18 keV
stigmatism -6.0 9.0 nm
working_distance -0.12 0.18 mm
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 SwinIR + gradient 0.661 26.38 0.842
2 Wiener Filter + gradient 0.484 19.27 0.563
3 BM3D + gradient 0.463 18.42 0.521
4 Noise2Void + gradient 0.441 17.75 0.488
Spec Ranges (3 parameters)
Parameter Min Max Unit
beam_energy -0.07 0.23 keV
stigmatism -3.5 11.5 nm
working_distance -0.07 0.23 mm
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

SEM forms images by rastering a focused electron beam (1-30 keV) across the specimen surface and collecting secondary electrons (SE, topographic contrast) or backscattered electrons (BSE, compositional Z-contrast). Resolution is determined by the probe diameter (1-10 nm), accelerating voltage, and interaction volume. Key artifacts include charging in non-conductive specimens, drift, and contamination.

Principle

Scanning Electron Microscopy rasters a focused electron beam (0.1-30 keV) across the sample surface. Secondary electrons (SE) emitted from the top few nanometers provide topographic contrast, while backscattered electrons (BSE) from deeper interactions reveal compositional contrast (higher Z → more BSE). The image is formed point-by-point, with resolution down to 1-5 nm determined by the probe size.

How to Build the System

Operate a field-emission SEM (FEG-SEM, e.g., Zeiss GeminiSEM, JEOL JSM-7800F) under high vacuum (< 10⁻⁴ Pa). Mount samples on conductive stubs with carbon tape or silver paint. Non-conductive samples must be sputter-coated (5-10 nm Au/Pd or C) to prevent charging. Set accelerating voltage (1-5 kV for surface detail, 10-20 kV for BSE compositional contrast). Select appropriate detectors (Everhart-Thornley for SE, solid-state for BSE). Align the column and perform astigmatism correction.

Common Reconstruction Algorithms

  • Noise reduction by frame averaging or Kalman filtering
  • Charging artifact compensation (dynamic focus, low-kV imaging)
  • 3-D surface reconstruction from stereo-pair SEM images
  • Deep-learning SEM denoising (for low-dose or fast-scan images)
  • Automated particle analysis and morphometry

Common Mistakes

  • Sample charging causing bright streaks and image distortion
  • Astigmatism not corrected, producing elongated features
  • Excessive beam current damaging or contaminating delicate samples
  • Carbon contamination from residual hydrocarbons in the chamber
  • Wrong working distance causing suboptimal resolution or depth of field

How to Avoid Mistakes

  • Coat non-conductive samples or use low-vacuum/variable-pressure mode
  • Correct astigmatism carefully using the wobbler on a recognizable feature
  • Use the minimum beam current needed; work at low kV for beam-sensitive samples
  • Plasma-clean the chamber and samples; use a cold trap to reduce contamination
  • Optimize working distance for the specific detector and resolution requirement

Forward-Model Mismatch Cases

  • The widefield fallback applies optical Gaussian blur, but SEM image formation involves electron-sample interaction (secondary electron yield depends on surface topography and composition) — the contrast mechanism is fundamentally different from optical fluorescence
  • SEM contrast (SE and BSE signals) depends on accelerating voltage, material Z-number, surface tilt, and detector geometry — the widefield PSF convolution model cannot capture these electron-matter interaction physics

How to Correct the Mismatch

  • Use the SEM operator that models the electron probe profile (sub-nm spot) and secondary/backscattered electron yield as a function of local surface topography and composition
  • Include the interaction volume (Monte Carlo electron trajectory simulation), detector angular acceptance, and signal mixing between SE (topography) and BSE (composition) channels

Experimental Setup — Signal Chain

Experimental setup diagram for Scanning Electron Microscopy

Experimental Setup

Instrument: JEOL JSM-7800F / Thermo Fisher Apreo 2 / Zeiss GeminiSEM 560
Accelerating Voltage Kv: 10
Beam Current Na: 0.54
Working Distance Mm: 10
Pixel Size Nm: 7.1
Magnification: 20,000x
Detector: Everhart-Thornley (SE2) + in-lens (SE1)
Image Size: 1024x768

Key References

  • Goldstein et al., 'Scanning Electron Microscopy and X-ray Microanalysis', Springer (2018)

Canonical Datasets

  • SEM Dataset for Nanomaterial Segmentation (Aversa et al.)
  • NIST SEM calibration images

Spec DAG — Forward Model Pipeline

P(e⁻ beam) → C(probe) → D(g, η₁)

P Electron Beam (e⁻)
C Probe Scanning (probe)
D SE / BSE Detector (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
ΔE beam_energy Beam energy error (keV) 0 0.1
ΔA_s stigmatism Astigmatism (nm) 0 5.0
ΔWD working_distance Working distance error (mm) 0 0.1

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.