SIM

Structured Illumination 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
🥇 SIMformer 0.838 36.5 0.960 ✓ Certified SIM reconstruction transformer, 2024
🥈 DL-SIM 0.806 35.0 0.945 ✓ Certified Jin et al., Nat. Methods 2023
🥉 PnP-SIM 0.720 31.5 0.890 ✓ Certified PnP with SIM forward model
4 Wiener-SIM 0.635 28.5 0.820 ✓ Certified Gustafsson, J. Microsc. 2000

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
🥇 SIMformer + gradient 0.773
0.805
34.37 dB / 0.964
0.791
33.78 dB / 0.959
0.723
29.38 dB / 0.907
✓ Certified SIM reconstruction transformer, 2024
🥈 DL-SIM + gradient 0.676
0.785
33.01 dB / 0.953
0.660
25.49 dB / 0.817
0.584
22.81 dB / 0.724
✓ Certified Jin et al., Nat. Methods 2023
🥉 PnP-SIM + gradient 0.660
0.731
29.3 dB / 0.906
0.648
25.21 dB / 0.809
0.600
24.01 dB / 0.769
✓ Certified PnP with SIM forward model
4 Wiener-SIM + gradient 0.658
0.698
26.85 dB / 0.854
0.662
26.24 dB / 0.839
0.615
23.97 dB / 0.767
✓ Certified Gustafsson, J. Microsc. 2000

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 SIMformer + gradient 0.805 34.37 0.964
2 DL-SIM + gradient 0.785 33.01 0.953
3 PnP-SIM + gradient 0.731 29.3 0.906
4 Wiener-SIM + gradient 0.698 26.85 0.854
Spec Ranges (3 parameters)
Parameter Min Max Unit
pattern_phase -0.05 0.1 rad
pattern_freq -1.0 2.0 %
modulation_depth -5.0 10.0 %
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 SIMformer + gradient 0.791 33.78 0.959
2 Wiener-SIM + gradient 0.662 26.24 0.839
3 DL-SIM + gradient 0.660 25.49 0.817
4 PnP-SIM + gradient 0.648 25.21 0.809
Spec Ranges (3 parameters)
Parameter Min Max Unit
pattern_phase -0.06 0.09 rad
pattern_freq -1.2 1.8 %
modulation_depth -6.0 9.0 %
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 SIMformer + gradient 0.723 29.38 0.907
2 Wiener-SIM + gradient 0.615 23.97 0.767
3 PnP-SIM + gradient 0.600 24.01 0.769
4 DL-SIM + gradient 0.584 22.81 0.724
Spec Ranges (3 parameters)
Parameter Min Max Unit
pattern_phase -0.035 0.115 rad
pattern_freq -0.7 2.3 %
modulation_depth -3.5 11.5 %
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

Structured illumination microscopy (SIM) achieves ~2x lateral resolution improvement by illuminating the sample with sinusoidal patterns at multiple orientations and phases. Frequency mixing between the illumination pattern and sample structure shifts high-frequency information into the microscope passband. Reconstruction separates and reassembles frequency components via Wiener-SIM or deep-learning SIM. The forward model is y_k = PSF ** (I_k * x) + n for each pattern k.

Principle

Structured Illumination Microscopy projects a known sinusoidal pattern onto the specimen, shifting high-frequency spatial information into the observable passband via Moiré interference. Multiple images (typically 9-15) are acquired at different pattern orientations and phases, then computationally recombined in Fourier space to achieve ~2× lateral resolution improvement beyond the diffraction limit.

How to Build the System

Install a SIM-capable microscope (Nikon N-SIM, Zeiss Elyra 7, or custom with SLM/DMD). Use a high-NA objective (100x 1.49 NA TIRF) for maximum frequency extension. The illumination grating (SLM or fiber interference) generates the sinusoidal pattern. Acquire 3 orientations × 3-5 phases. A fast sCMOS camera captures all raw frames in ~100-500 ms for 2D-SIM. Careful alignment of the pattern contrast is critical.

Common Reconstruction Algorithms

  • Gustafsson/Heintzmann frequency-domain SIM reconstruction
  • Open-source fairSIM (ImageJ plugin)
  • Wiener-filtered order separation and recombination
  • Deep-learning SIM (ML-SIM, reconstruction from fewer frames)
  • Hessian-SIM for live-cell with reduced artifacts

Common Mistakes

  • Insufficient pattern contrast causing weak Moiré fringes and honeycomb artifacts
  • Misaligned illumination orders producing stripe artifacts in the reconstruction
  • Over-processing (too aggressive Wiener parameter) creating ringing artifacts
  • Using objectives with insufficient NA for the desired resolution gain
  • Photobleaching between pattern acquisitions causing intensity inconsistency

How to Avoid Mistakes

  • Verify pattern contrast >0.5 on a thin uniform fluorescent layer before experiments
  • Calibrate illumination pattern positions/angles using SIMcheck (ImageJ plugin)
  • Tune the Wiener parameter conservatively; use SIMcheck to assess reconstruction quality
  • Use 1.49 NA objectives for maximum resolution; 1.40 NA limits SIM performance
  • Minimize total acquisition time; use fast cameras and short exposures

Forward-Model Mismatch Cases

  • The widefield fallback produces a single (64,64) blurred image, but SIM requires 9-15 raw frames (3 orientations x 3-5 phases) with structured illumination patterns — output shape (64,64,9) vs (64,64)
  • Without the sinusoidal illumination pattern encoding, the high-frequency information that SIM moves into the passband via Moiré interference is completely absent — no super-resolution is possible

How to Correct the Mismatch

  • Use the SIM operator that generates multiple pattern-modulated images: y_k = (1 + m*cos(k_i*r + phi_j)) * (PSF ** x) for each orientation i and phase j
  • Reconstruct using Fourier-space order separation and recombination (Gustafsson method) or deep-learning SIM, which require the correct multi-frame structured illumination forward model

Experimental Setup — Signal Chain

Experimental setup diagram for Structured Illumination Microscopy

Experimental Setup

Instrument: Zeiss Elyra 7 / Nikon N-SIM S
Objective: Apo TIRF 100x / 1.49 NA oil
Pixel Size Nm: 32
Excitation Source: 488 nm laser (20 mW)
Orientations: 3
Phases Per Orientation: 5
Raw Images: 15
Achieved Resolution Nm: 110
Detector: Hamamatsu ORCA-Flash4.0 sCMOS
Pattern Generator: SLM / diffraction grating
Reconstruction: Wiener-SIM / fairSIM

Key References

  • Gustafsson, 'Surpassing the lateral resolution limit by a factor of two using structured illumination microscopy', J. Microsc. 198, 82-87 (2000)
  • Muller & Bhatt, 'Open-source image reconstruction of super-resolution structured illumination microscopy data (fairSIM)', Nature Comms 7, 10980 (2016)

Canonical Datasets

  • BioSR SIM paired dataset (Zhang et al., Nature Methods 2023)
  • fairSIM test datasets (Hagen et al.)

Spec DAG — Forward Model Pipeline

S(grating) → C(PSF) → Σ_φ → D(g, η₃)

S Sinusoidal Illumination (grating)
C PSF Convolution (PSF)
Σ Phase-Shift Sum (φ)
D sCMOS Camera (g, η₃)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δφ pattern_phase Pattern phase error (rad) 0 0.05
Δk pattern_freq Pattern frequency error (%) 0 1.0
Δm modulation_depth Modulation depth error (%) 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.