Correlative Light-Electron Microscopy (CLEM)

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
🥇	DiffusionCLEM DiffusionCLEM Chen 2024 39.1 dB SSIM 0.958 Checkpoint unavailable	0.881	39.1	0.958	✓ Certified	Chen 2024
🥈	SwinCLEM SwinCLEM Huang 2023 37.5 dB SSIM 0.944 Checkpoint unavailable	0.847	37.5	0.944	✓ Certified	Huang 2023
🥉	TransMorph TransMorph Chen 2022 36.2 dB SSIM 0.931 Checkpoint unavailable	0.819	36.2	0.931	✓ Certified	Chen 2022
4	PINN-CLEM PINN-CLEM Löffler 2023 35.8 dB SSIM 0.927 Checkpoint unavailable	0.810	35.8	0.927	✓ Certified	Löffler 2023
5	CLEM-Net CLEM-Net Spiers 2021 34.5 dB SSIM 0.912 Checkpoint unavailable	0.781	34.5	0.912	✓ Certified	Spiers 2021
6	VoxelMorph VoxelMorph Balakrishnan 2019 32.8 dB SSIM 0.890 Checkpoint unavailable	0.742	32.8	0.890	✓ Certified	Balakrishnan 2019
7	CNN-Reg CNN-Reg de Vos 2019 30.2 dB SSIM 0.855 Checkpoint unavailable	0.681	30.2	0.855	✓ Certified	de Vos 2019
8	Landmark-Reg Landmark-Reg Arganda-Carreras 2006 25.8 dB SSIM 0.782 Try in SpecLab →	0.571	25.8	0.782	✓ Certified	Arganda-Carreras 2006
9	Cross-Correlation Cross-Correlation Thévenaz 1998 23.5 dB SSIM 0.741 Try in SpecLab →	0.512	23.5	0.741	✓ Certified	Thévenaz 1998

Dataset: PWM Benchmark (9 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
🥇	SwinCLEM + gradient SwinCLEM + gradient Huang et al., IEEE TMI 2023 Score 0.773 Correct & Reconstruct →	0.773	0.817 35.31 dB / 0.970	0.775 31.89 dB / 0.941	0.726 28.98 dB / 0.900	✓ Certified	Huang et al., IEEE TMI 2023
🥈	DiffusionCLEM + gradient DiffusionCLEM + gradient Chen et al., Nat. Methods 2024 Score 0.771 Correct & Reconstruct →	0.771	0.834 36.37 dB / 0.975	0.765 31.81 dB / 0.941	0.715 28.59 dB / 0.893	✓ Certified	Chen et al., Nat. Methods 2024
🥉	TransMorph + gradient TransMorph + gradient Chen et al., Med. Image Anal. 2022 Score 0.740 Correct & Reconstruct →	0.740	0.821 34.67 dB / 0.966	0.733 29.52 dB / 0.909	0.667 26.18 dB / 0.837	✓ Certified	Chen et al., Med. Image Anal. 2022
4	CLEM-Net + gradient CLEM-Net + gradient Spiers et al., Nat. Methods 2021 Score 0.716 Correct & Reconstruct →	0.716	0.779 32.38 dB / 0.947	0.704 27.66 dB / 0.873	0.666 26.74 dB / 0.852	✓ Certified	Spiers et al., Nat. Methods 2021
5	VoxelMorph + gradient VoxelMorph + gradient Balakrishnan et al., IEEE TPAMI 2019 Score 0.693 Correct & Reconstruct →	0.693	0.774 31.38 dB / 0.936	0.691 27.5 dB / 0.870	0.615 24.07 dB / 0.771	✓ Certified	Balakrishnan et al., IEEE TPAMI 2019
6	PINN-CLEM + gradient PINN-CLEM + gradient Löffler et al., Nat. Methods 2023 Score 0.688 Correct & Reconstruct →	0.688	0.818 34.8 dB / 0.966	0.678 26.94 dB / 0.857	0.568 22.14 dB / 0.696	✓ Certified	Löffler et al., Nat. Methods 2023
7	CNN-Reg + gradient CNN-Reg + gradient de Vos et al., NeuroImage 2019 Score 0.600 Correct & Reconstruct →	0.600	0.712 28.31 dB / 0.887	0.579 22.25 dB / 0.701	0.509 19.99 dB / 0.598	✓ Certified	de Vos et al., NeuroImage 2019
8	Landmark-Reg + gradient Landmark-Reg + gradient Arganda-Carreras et al., Bioinformatics 2006 Score 0.585 Correct & Reconstruct →	0.585	0.604 23.05 dB / 0.733	0.590 23.44 dB / 0.748	0.560 22.6 dB / 0.715	✓ Certified	Arganda-Carreras et al., Bioinformatics 2006
9	Cross-Correlation + gradient Cross-Correlation + gradient Thévenaz et al., IEEE TIP 1998 Score 0.509 Correct & Reconstruct →	0.509	0.593 22.48 dB / 0.710	0.492 19.82 dB / 0.590	0.441 18.43 dB / 0.521	✓ Certified	Thévenaz et al., IEEE TIP 1998

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	DiffusionCLEM + gradient	0.834	36.37	0.975
2	TransMorph + gradient	0.821	34.67	0.966
3	PINN-CLEM + gradient	0.818	34.8	0.966
4	SwinCLEM + gradient	0.817	35.31	0.97
5	CLEM-Net + gradient	0.779	32.38	0.947
6	VoxelMorph + gradient	0.774	31.38	0.936
7	CNN-Reg + gradient	0.712	28.31	0.887
8	Landmark-Reg + gradient	0.604	23.05	0.733
9	Cross-Correlation + gradient	0.593	22.48	0.71

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
registration_error_(lm_to_em)	-100.0	200.0	nm
sample_deformation_(fixation)	-1.0	2.0	shrinkage
fluorescence_preservation	72.0	114.0	-

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	SwinCLEM + gradient	0.775	31.89	0.941
2	DiffusionCLEM + gradient	0.765	31.81	0.941
3	TransMorph + gradient	0.733	29.52	0.909
4	CLEM-Net + gradient	0.704	27.66	0.873
5	VoxelMorph + gradient	0.691	27.5	0.87
6	PINN-CLEM + gradient	0.678	26.94	0.857
7	Landmark-Reg + gradient	0.590	23.44	0.748
8	CNN-Reg + gradient	0.579	22.25	0.701
9	Cross-Correlation + gradient	0.492	19.82	0.59

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
registration_error_(lm_to_em)	-120.0	180.0	nm
sample_deformation_(fixation)	-1.2	1.8	shrinkage
fluorescence_preservation	74.8	116.8	-

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	SwinCLEM + gradient	0.726	28.98	0.9
2	DiffusionCLEM + gradient	0.715	28.59	0.893
3	TransMorph + gradient	0.667	26.18	0.837
4	CLEM-Net + gradient	0.666	26.74	0.852
5	VoxelMorph + gradient	0.615	24.07	0.771
6	PINN-CLEM + gradient	0.568	22.14	0.696
7	Landmark-Reg + gradient	0.560	22.6	0.715
8	CNN-Reg + gradient	0.509	19.99	0.598
9	Cross-Correlation + gradient	0.441	18.43	0.521

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
registration_error_(lm_to_em)	-70.0	230.0	nm
sample_deformation_(fixation)	-0.7	2.3	shrinkage
fluorescence_preservation	67.8	109.8	-

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

Spec DAG — Forward Model Pipeline

(C → D) + (C → D) → ⊕

C Convolution

D Detector (LM)

C Convolution

D Detector (EM)

Fusion Fusion

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
r_e	registration_error_(lm_to_em)	Registration error (LM to EM) (nm)	0.0	100.0
s_d	sample_deformation_(fixation)	Sample deformation (fixation) (shrinkage)	0.0	1.0
f_p	fluorescence_preservation	Fluorescence preservation (-)	100.0	86.0

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.