Light-Sheet

Light-Sheet Fluorescence 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
🥇 ScoreMicro 0.882 38.48 0.981 ✓ Certified Wei et al., ECCV 2025
🥈 DiffDeconv 0.875 38.12 0.979 ✓ Certified Huang et al., NeurIPS 2024
🥉 Restormer+ 0.865 37.65 0.975 ✓ Certified Zamir et al., ICCV 2024
4 DeconvFormer 0.857 37.25 0.972 ✓ Certified Chen et al., CVPR 2024
5 ResUNet 0.830 35.85 0.964 ✓ Certified DeCelle et al., Nat. Methods 2021
6 Restormer 0.828 35.8 0.962 ✓ Certified Zamir et al., CVPR 2022
7 U-Net 0.814 35.15 0.956 ✓ Certified Ronneberger et al., MICCAI 2015
8 CARE 0.799 34.5 0.948 ✓ Certified Weigert et al., Nat. Methods 2018
9 PnP-DnCNN 0.715 31.2 0.890 ✓ Certified Zhang et al., IEEE TIP 2017
10 PnP-FISTA 0.693 30.42 0.872 ✓ Certified Bai et al., 2020
11 TV-Deconvolution 0.664 29.5 0.845 ✓ Certified TV-regularized deconvolution
12 Wiener Filter 0.625 28.35 0.805 ✓ Certified Analytical baseline
13 Richardson-Lucy 0.587 27.1 0.770 ✓ Certified Richardson 1972 / Lucy 1974

Dataset: PWM Benchmark (13 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
🥇 Restormer+ + gradient 0.785
0.838
36.04 dB / 0.974
0.771
33.17 dB / 0.954
0.746
31.11 dB / 0.932
✓ Certified Zamir et al., ICCV 2024
🥈 ScoreMicro + gradient 0.780
0.848
36.73 dB / 0.977
0.758
31.1 dB / 0.932
0.733
30.4 dB / 0.923
✓ Certified Wei et al., ECCV 2025
🥉 Restormer + gradient 0.773
0.818
34.77 dB / 0.966
0.769
31.55 dB / 0.938
0.733
30.38 dB / 0.922
✓ Certified Zamir et al., CVPR 2022
4 DeconvFormer + gradient 0.772
0.834
36.06 dB / 0.974
0.772
32.8 dB / 0.951
0.710
29.7 dB / 0.912
✓ Certified Chen et al., CVPR 2024
5 DiffDeconv + gradient 0.754
0.822
35.58 dB / 0.971
0.734
30.32 dB / 0.922
0.705
28.59 dB / 0.893
✓ Certified Huang et al., NeurIPS 2024
6 ResUNet + gradient 0.726
0.794
33.36 dB / 0.956
0.722
29.19 dB / 0.904
0.662
26.14 dB / 0.836
✓ Certified DeCelle et al., Nat. Methods 2021
7 U-Net + gradient 0.720
0.806
33.52 dB / 0.957
0.697
28.25 dB / 0.886
0.656
26.43 dB / 0.844
✓ Certified Ronneberger et al., MICCAI 2015
8 CARE + gradient 0.710
0.778
32.35 dB / 0.946
0.700
27.55 dB / 0.871
0.653
26.28 dB / 0.840
✓ Certified Weigert et al., Nat. Methods 2018
9 PnP-DnCNN + gradient 0.700
0.748
29.56 dB / 0.910
0.682
26.65 dB / 0.849
0.670
26.62 dB / 0.849
✓ Certified Zhang et al., IEEE TIP 2017
10 TV-Deconvolution + gradient 0.677
0.719
27.81 dB / 0.877
0.663
26.67 dB / 0.850
0.649
26.12 dB / 0.835
✓ Certified Rudin et al., Phys. A 1992
11 PnP-FISTA + gradient 0.643
0.736
28.75 dB / 0.896
0.629
24.21 dB / 0.776
0.564
22.44 dB / 0.708
✓ Certified Bai et al., 2020
12 Wiener Filter + gradient 0.625
0.662
25.43 dB / 0.815
0.606
23.44 dB / 0.748
0.606
23.87 dB / 0.764
✓ Certified Analytical baseline
13 Richardson-Lucy + gradient 0.607
0.650
25.3 dB / 0.812
0.599
23.17 dB / 0.738
0.572
22.13 dB / 0.696
✓ Certified Richardson, JOSA 1972 / Lucy, AJ 1974

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 ScoreMicro + gradient 0.848 36.73 0.977
2 Restormer+ + gradient 0.838 36.04 0.974
3 DeconvFormer + gradient 0.834 36.06 0.974
4 DiffDeconv + gradient 0.822 35.58 0.971
5 Restormer + gradient 0.818 34.77 0.966
6 U-Net + gradient 0.806 33.52 0.957
7 ResUNet + gradient 0.794 33.36 0.956
8 CARE + gradient 0.778 32.35 0.946
9 PnP-DnCNN + gradient 0.748 29.56 0.91
10 PnP-FISTA + gradient 0.736 28.75 0.896
11 TV-Deconvolution + gradient 0.719 27.81 0.877
12 Wiener Filter + gradient 0.662 25.43 0.815
13 Richardson-Lucy + gradient 0.650 25.3 0.812
Spec Ranges (3 parameters)
Parameter Min Max Unit
sheet_thickness -1.0 2.0 μm
sheet_tilt -0.5 1.0 deg
stripe_artifact -0.1 0.2
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 DeconvFormer + gradient 0.772 32.8 0.951
2 Restormer+ + gradient 0.771 33.17 0.954
3 Restormer + gradient 0.769 31.55 0.938
4 ScoreMicro + gradient 0.758 31.1 0.932
5 DiffDeconv + gradient 0.734 30.32 0.922
6 ResUNet + gradient 0.722 29.19 0.904
7 CARE + gradient 0.700 27.55 0.871
8 U-Net + gradient 0.697 28.25 0.886
9 PnP-DnCNN + gradient 0.682 26.65 0.849
10 TV-Deconvolution + gradient 0.663 26.67 0.85
11 PnP-FISTA + gradient 0.629 24.21 0.776
12 Wiener Filter + gradient 0.606 23.44 0.748
13 Richardson-Lucy + gradient 0.599 23.17 0.738
Spec Ranges (3 parameters)
Parameter Min Max Unit
sheet_thickness -1.2 1.8 μm
sheet_tilt -0.6 0.9 deg
stripe_artifact -0.12 0.18
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 Restormer+ + gradient 0.746 31.11 0.932
2 ScoreMicro + gradient 0.733 30.4 0.923
3 Restormer + gradient 0.733 30.38 0.922
4 DeconvFormer + gradient 0.710 29.7 0.912
5 DiffDeconv + gradient 0.705 28.59 0.893
6 PnP-DnCNN + gradient 0.670 26.62 0.849
7 ResUNet + gradient 0.662 26.14 0.836
8 U-Net + gradient 0.656 26.43 0.844
9 CARE + gradient 0.653 26.28 0.84
10 TV-Deconvolution + gradient 0.649 26.12 0.835
11 Wiener Filter + gradient 0.606 23.87 0.764
12 Richardson-Lucy + gradient 0.572 22.13 0.696
13 PnP-FISTA + gradient 0.564 22.44 0.708
Spec Ranges (3 parameters)
Parameter Min Max Unit
sheet_thickness -0.7 2.3 μm
sheet_tilt -0.35 1.15 deg
stripe_artifact -0.07 0.23
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

Light-sheet microscopy (LSFM / SPIM) illuminates the sample with a thin sheet of light perpendicular to the detection axis, providing intrinsic optical sectioning. Primary artifacts are stripe patterns caused by absorption and scattering in the illumination path, plus anisotropic PSF blur. The forward model is y = S(z) * (PSF_3d *** x) + n where S(z) models the stripe attenuation. Reconstruction involves destriping followed by optional deconvolution.

Principle

A thin sheet of laser light illuminates only the focal plane of the detection objective, providing intrinsic optical sectioning with minimal out-of-plane photobleaching. The orthogonal geometry between illumination and detection decouples sectioning from resolution. Detection is widefield, enabling fast volumetric imaging of large specimens.

How to Build the System

Arrange two orthogonal objective arms: one for the excitation sheet (cylindrical lens or digitally scanned Gaussian/Bessel beam) and one for detection (high-NA water-dipping). Mount the sample in agarose or hold in a chamber compatible with the dual-objective geometry. Use a fast sCMOS camera for detection. Stage scanning or sheet scanning acquires z-stacks. Consider diSPIM (dual-view) for isotropic resolution.

Common Reconstruction Algorithms

  • Multi-view fusion (weighted averaging of complementary views)
  • Multi-view deconvolution (Bayesian, joint Richardson-Lucy)
  • Content-based image fusion
  • Deep-learning denoising for high-speed acquisitions (CARE)
  • Stripe artifact removal (wavelet-FFT filtering)

Common Mistakes

  • Light sheet too thick, degrading axial resolution and sectioning
  • Absorption and scattering in thick tissue causing shadow artifacts (stripes)
  • Misalignment between sheet focal plane and detection focal plane
  • Improper sample mounting causing drift or deformation during long acquisitions
  • Ignoring refractive-index variations causing sheet deflection inside tissue

How to Avoid Mistakes

  • Use Bessel or lattice light sheet for thin, uniform illumination profiles
  • Pivot the light sheet or use dual-side illumination to reduce shadow artifacts
  • Carefully co-align illumination and detection planes using fluorescent beads
  • Use stable, low-melting-point agarose embedding and vibration-isolated stages
  • Clear or match refractive index of tissue where possible; use adaptive optics

Forward-Model Mismatch Cases

  • The widefield fallback processes only 2D (64,64) images, but light-sheet microscopy acquires 3D volumes (64,64,32) with intrinsic optical sectioning — the volumetric z-dimension is entirely lost
  • Widefield illumination excites the entire sample volume causing out-of-focus blur, whereas the light sheet illuminates only the focal plane — the fallback forward model includes fluorescence contributions from planes that the real system never excites

How to Correct the Mismatch

  • Use the lightsheet operator that processes 3D volumes with the sheet illumination profile: each z-slice is excited only by the thin (1-5 um) light sheet
  • Model the sheet thickness and propagation (Gaussian or Bessel beam) explicitly; for multi-view systems, include the detection PSF from the orthogonal objective

Experimental Setup — Signal Chain

Experimental setup diagram for Light-Sheet Fluorescence Microscopy

Experimental Setup

Instrument: Zeiss Lightsheet 7 / LaVision BioTec UltraMicroscope II
Detection Objective: Plan Apo 20x / 1.0 NA water dipping
Illumination Na: 0.1
Pixel Size Nm: 406
Sheet Thickness Um: 5
Excitation Source: 488 nm laser (10 mW)
Frame Rate Fps: 10
Sample: zebrafish embryo / cleared tissue
Detector: Hamamatsu ORCA-Flash4.0 sCMOS
Reconstruction: deconvolution + destriping

Key References

  • Huisken et al., 'Optical sectioning deep inside live embryos by SPIM', Science 305, 1007-1009 (2004)
  • Power & Bhatt, 'A guide to light-sheet fluorescence microscopy for multiscale imaging', Nature Methods 14, 360-373 (2017)

Canonical Datasets

  • OpenSPIM sample datasets
  • Zebrafish developmental lightsheet atlas

Spec DAG — Forward Model Pipeline

C(PSF_sheet) → D(g, η₃)

C Light-Sheet PSF (PSF_sheet)
D sCMOS Camera (g, η₃)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δw sheet_thickness Sheet thickness error (μm) 0 1.0
Δθ sheet_tilt Sheet tilt (deg) 0 0.5
α_s stripe_artifact Stripe artifact amplitude 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.