Expansion Microscopy (ExM)

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
🥇	DiffExM DiffExM Gao et al. 2024 40.0 dB SSIM 0.960 Checkpoint unavailable	0.897	40.0	0.960	✓ Certified	Gao et al. 2024
🥈	PhysExM PhysExM Chen et al. 2024 38.8 dB SSIM 0.950 Checkpoint unavailable	0.872	38.8	0.950	✓ Certified	Chen et al. 2024
🥉	SwinExM SwinExM Wang et al. 2023 37.7 dB SSIM 0.941 Checkpoint unavailable	0.849	37.7	0.941	✓ Certified	Wang et al. 2023
4	TransExM TransExM Li et al. 2022 36.3 dB SSIM 0.927 Checkpoint unavailable	0.819	36.3	0.927	✓ Certified	Li et al. 2022
5	DeepInterp-ExM DeepInterp-ExM Lecoq et al. 2021 34.2 dB SSIM 0.898 Checkpoint unavailable	0.769	34.2	0.898	✓ Certified	Lecoq et al. 2021
6	DnCNN-ExM DnCNN-ExM Zhao et al. 2019 31.8 dB SSIM 0.860 Checkpoint unavailable	0.710	31.8	0.860	✓ Certified	Zhao et al. 2019
7	TV-ExM TV-ExM Rudin et al. 1992 29.1 dB SSIM 0.819 Try in SpecLab →	0.644	29.1	0.819	✓ Certified	Rudin et al. 1992
8	RL-ExM RL-ExM Richardson 1972 26.9 dB SSIM 0.778 Try in SpecLab →	0.587	26.9	0.778	✓ Certified	Richardson 1972
9	Deconv-Exp Deconv-Exp Chen et al. 2015 24.5 dB SSIM 0.742 Try in SpecLab →	0.529	24.5	0.742	✓ Certified	Chen et al. 2015

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
🥇	DiffExM + gradient DiffExM + gradient Gao et al., NeurIPS 2024 Score 0.784 Correct & Reconstruct →	0.784	0.846 37.23 dB / 0.979	0.765 31.43 dB / 0.936	0.741 29.86 dB / 0.915	✓ Certified	Gao et al., NeurIPS 2024
🥈	SwinExM + gradient SwinExM + gradient Wang et al., Cell Syst. 2023 Score 0.773 Correct & Reconstruct →	0.773	0.817 34.99 dB / 0.968	0.760 32.33 dB / 0.946	0.741 29.88 dB / 0.915	✓ Certified	Wang et al., Cell Syst. 2023
🥉	TransExM + gradient TransExM + gradient Li et al., Nat. Methods 2022 Score 0.764 Correct & Reconstruct →	0.764	0.822 34.6 dB / 0.965	0.741 31.02 dB / 0.931	0.728 29.95 dB / 0.916	✓ Certified	Li et al., Nat. Methods 2022
4	PhysExM + gradient PhysExM + gradient Chen et al., Nat. Commun. 2024 Score 0.762 Correct & Reconstruct →	0.762	0.831 35.95 dB / 0.973	0.758 31.53 dB / 0.937	0.696 28.42 dB / 0.889	✓ Certified	Chen et al., Nat. Commun. 2024
5	DnCNN-ExM + gradient DnCNN-ExM + gradient Zhao et al., Nat. Methods 2019 Score 0.664 Correct & Reconstruct →	0.664	0.735 29.26 dB / 0.905	0.650 25.68 dB / 0.823	0.606 23.45 dB / 0.748	✓ Certified	Zhao et al., Nat. Methods 2019
6	DeepInterp-ExM + gradient DeepInterp-ExM + gradient Lecoq et al., Nat. Methods 2021 Score 0.645 Correct & Reconstruct →	0.645	0.771 31.72 dB / 0.940	0.632 24.7 dB / 0.792	0.531 21.46 dB / 0.666	✓ Certified	Lecoq et al., Nat. Methods 2021
7	RL-ExM + gradient RL-ExM + gradient Richardson, J. Opt. Soc. Am. 1972 Score 0.605 Correct & Reconstruct →	0.605	0.637 24.5 dB / 0.786	0.602 23.86 dB / 0.764	0.577 23.09 dB / 0.735	✓ Certified	Richardson, J. Opt. Soc. Am. 1972
8	Deconv-Exp + gradient Deconv-Exp + gradient Chen et al., Science 2015 Score 0.536 Correct & Reconstruct →	0.536	0.573 21.89 dB / 0.685	0.554 21.72 dB / 0.678	0.481 19.69 dB / 0.584	✓ Certified	Chen et al., Science 2015
9	TV-ExM + gradient TV-ExM + gradient Rudin et al., Physica D 1992 Score 0.520 Correct & Reconstruct →	0.520	0.710 27.37 dB / 0.867	0.455 18.03 dB / 0.501	0.394 16.6 dB / 0.430	✓ Certified	Rudin et al., Physica D 1992

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	DiffExM + gradient	0.846	37.23	0.979
2	PhysExM + gradient	0.831	35.95	0.973
3	TransExM + gradient	0.822	34.6	0.965
4	SwinExM + gradient	0.817	34.99	0.968
5	DeepInterp-ExM + gradient	0.771	31.72	0.94
6	DnCNN-ExM + gradient	0.735	29.26	0.905
7	TV-ExM + gradient	0.710	27.37	0.867
8	RL-ExM + gradient	0.637	24.5	0.786
9	Deconv-Exp + gradient	0.573	21.89	0.685

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
expansion_factor	3.9	4.2	x
local_distortion	-1.0	2.0	relative
anisotropic_expansion	-0.6	1.2	xvsy

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	DiffExM + gradient	0.765	31.43	0.936
2	SwinExM + gradient	0.760	32.33	0.946
3	PhysExM + gradient	0.758	31.53	0.937
4	TransExM + gradient	0.741	31.02	0.931
5	DnCNN-ExM + gradient	0.650	25.68	0.823
6	DeepInterp-ExM + gradient	0.632	24.7	0.792
7	RL-ExM + gradient	0.602	23.86	0.764
8	Deconv-Exp + gradient	0.554	21.72	0.678
9	TV-ExM + gradient	0.455	18.03	0.501

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
expansion_factor	3.88	4.18	x
local_distortion	-1.2	1.8	relative
anisotropic_expansion	-0.72	1.08	xvsy

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	DiffExM + gradient	0.741	29.86	0.915
2	SwinExM + gradient	0.741	29.88	0.915
3	TransExM + gradient	0.728	29.95	0.916
4	PhysExM + gradient	0.696	28.42	0.889
5	DnCNN-ExM + gradient	0.606	23.45	0.748
6	RL-ExM + gradient	0.577	23.09	0.735
7	DeepInterp-ExM + gradient	0.531	21.46	0.666
8	Deconv-Exp + gradient	0.481	19.69	0.584
9	TV-ExM + gradient	0.394	16.6	0.43

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
expansion_factor	3.93	4.23	x
local_distortion	-0.7	2.3	relative
anisotropic_expansion	-0.42	1.38	xvsy

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 Convolution

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
e_f	expansion_factor	Expansion factor (x)	4.0	4.1
l_d	local_distortion	Local distortion (relative)	0.0	1.0
a_e	anisotropic_expansion	Anisotropic expansion (x vs y)	0.0	0.6

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.