TEM

Transmission 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 0.743
0.787
32.38 dB / 0.947
0.745
30.61 dB / 0.926
0.697
27.5 dB / 0.870
✓ Certified Liang et al., ICCVW 2021
🥈 BM3D + gradient 0.626
0.677
26.39 dB / 0.843
0.620
23.86 dB / 0.764
0.580
23.37 dB / 0.745
✓ Certified Dabov et al., IEEE TIP 2007
🥉 Noise2Void + gradient 0.600
0.756
30.09 dB / 0.918
0.581
22.77 dB / 0.722
0.462
18.47 dB / 0.523
✓ Certified Krull et al., CVPR 2019
4 Wiener Filter + gradient 0.576
0.582
22.3 dB / 0.703
0.579
22.87 dB / 0.726
0.566
21.87 dB / 0.684
✓ 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 0.787 32.38 0.947
2 Noise2Void + gradient 0.756 30.09 0.918
3 BM3D + gradient 0.677 26.39 0.843
4 Wiener Filter + gradient 0.582 22.3 0.703
Spec Ranges (3 parameters)
Parameter Min Max Unit
defocus -50.0 100.0 nm
Cs -0.01 0.02 mm
beam_tilt -0.5 1.0 mrad
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 0.745 30.61 0.926
2 BM3D + gradient 0.620 23.86 0.764
3 Noise2Void + gradient 0.581 22.77 0.722
4 Wiener Filter + gradient 0.579 22.87 0.726
Spec Ranges (3 parameters)
Parameter Min Max Unit
defocus -60.0 90.0 nm
Cs -0.012 0.018 mm
beam_tilt -0.6 0.9 mrad
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 0.697 27.5 0.87
2 BM3D + gradient 0.580 23.37 0.745
3 Wiener Filter + gradient 0.566 21.87 0.684
4 Noise2Void + gradient 0.462 18.47 0.523
Spec Ranges (3 parameters)
Parameter Min Max Unit
defocus -35.0 115.0 nm
Cs -0.007 0.023 mm
beam_tilt -0.35 1.15 mrad
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

TEM transmits a high-energy electron beam (80-300 keV) through an ultra-thin specimen (<100 nm), magnifying the exit wave with EM lenses. In HRTEM, the image records interference between direct and diffracted beams, convolved by the contrast transfer function (CTF). The CTF introduces oscillating contrast reversals modulated by defocus and spherical aberration. Reconstruction involves CTF correction and, for biological specimens, single-particle averaging.

Principle

Transmission Electron Microscopy transmits a high-energy electron beam (80-300 keV) through an ultra-thin specimen (<100 nm). Electrons interact with the sample via elastic scattering (diffraction contrast, phase contrast) and inelastic scattering (energy loss). The transmitted beam is magnified by electromagnetic lenses to form an image with atomic-level resolution (0.05-0.2 nm in aberration-corrected TEMs).

How to Build the System

Operate a TEM (e.g., JEOL JEM-2100, Thermo Fisher Talos/Titan) under high vacuum (< 10⁻⁵ Pa). Prepare ultra-thin specimens using ultramicrotomy (biological), focused ion beam (FIB) milling (materials), or electropolishing (metals). Load samples on 3 mm TEM grids (Cu or Mo). Align the beam, correct condenser and objective astigmatism, and set appropriate defocus for phase contrast imaging. Use direct-electron detectors for highest DQE.

Common Reconstruction Algorithms

  • CTF correction (Contrast Transfer Function for phase contrast imaging)
  • Single-particle analysis (cryo-EM: classification, 3-D reconstruction)
  • Selected-area electron diffraction (SAED) pattern analysis
  • HRTEM image simulation (multislice or Bloch wave)
  • Deep-learning denoising for low-dose cryo-EM (Topaz, Warp, cryoSPARC)

Common Mistakes

  • Specimen too thick, causing multiple scattering and loss of interpretable contrast
  • Beam damage to organic or beam-sensitive materials from excessive electron dose
  • Astigmatism and coma not corrected, degrading high-resolution images
  • Not accounting for CTF effects when interpreting HRTEM images
  • Contamination building up on the specimen under the beam (hydrocarbon deposition)

How to Avoid Mistakes

  • Prepare specimens to <50 nm thickness; verify with EELS log-ratio thickness mapping
  • Use low-dose protocols and cryo-cooling for beam-sensitive specimens
  • Perform careful alignment including Zemlin tableau for Cs-corrected instruments
  • Simulate TEM images with known structure and compare; always correct CTF in analysis
  • Plasma-clean grids and specimens before loading; use a cryo-shield during imaging

Forward-Model Mismatch Cases

  • The widefield fallback produces real-valued output, but TEM forms images from coherent electron wave transmission — the complex-valued exit wave (amplitude and phase from elastic scattering) is lost, destroying quantitative phase-contrast information
  • TEM image contrast arises from coherent interference of scattered electron waves modulated by the contrast transfer function (CTF) — the widefield intensity-based Gaussian blur cannot model the oscillating CTF that produces Thon rings

How to Correct the Mismatch

  • Use the TEM operator that models coherent electron imaging: exit wave convolved with the CTF (including defocus, spherical aberration Cs, partial coherence) producing complex-valued image wave
  • Reconstruct phase and amplitude using CTF correction (Wiener filtering in Fourier space), or through-focus series exit-wave reconstruction for aberration-corrected quantitative HRTEM

Experimental Setup — Signal Chain

Experimental setup diagram for Transmission Electron Microscopy

Experimental Setup

Instrument: Thermo Fisher Titan Themis 300 / JEOL JEM-ARM300F2
Accelerating Voltage Kv: 300
Cs Corrected: True
Information Limit Pm: 50
Detector: Gatan K3 direct electron (5760x4092)
Pixel Size Pm: 50
Dose E Per A2: 30
Magnification: 1,000,000x

Key References

  • Williams & Carter, 'Transmission Electron Microscopy', Springer (2009)
  • Haider et al., 'Electron microscopy image enhanced', Nature 392, 768 (1998)

Canonical Datasets

  • EMPIAR (Electron Microscopy Public Image Archive)
  • NCEM atomic-resolution HRTEM benchmarks

Spec DAG — Forward Model Pipeline

P(e⁻) → C(CTF) → D(g, η₁)

P Electron Wave (e⁻)
C Contrast Transfer Function (CTF)
D Direct Electron Detector (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δf defocus Defocus error (nm) 0 50
ΔC_s Cs Spherical aberration error (mm) 0 0.01
Δθ beam_tilt Beam tilt error (mrad) 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.