Dark-Field 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
🥇	DiffusionDF DiffusionDF Luo 2023 40.3 dB SSIM 0.956 Checkpoint unavailable	0.900	40.3	0.956	✓ Certified	Luo 2023
🥈	Restormer-DF Restormer-DF Zamir 2022 38.9 dB SSIM 0.943 Checkpoint unavailable	0.870	38.9	0.943	✓ Certified	Zamir 2022
🥉	SwinIR-DF SwinIR-DF Liang 2021 37.6 dB SSIM 0.932 Checkpoint unavailable	0.843	37.6	0.932	✓ Certified	Liang 2021
4	CARE-DF CARE-DF Weigert 2018 35.1 dB SSIM 0.908 Checkpoint unavailable	0.789	35.1	0.908	✓ Certified	Weigert 2018
5	Noise2Void-DF Noise2Void-DF Krull 2019 33.7 dB SSIM 0.889 Checkpoint unavailable	0.756	33.7	0.889	✓ Certified	Krull 2019
6	BM3D-DF BM3D-DF Dabov 2007 32.4 dB SSIM 0.871 Try in SpecLab →	0.726	32.4	0.871	✓ Certified	Dabov 2007
7	TV-DF TV-DF Rudin 1992 29.8 dB SSIM 0.836 Try in SpecLab →	0.665	29.8	0.836	✓ Certified	Rudin 1992
8	Wiener-DF Wiener-DF Wiener 1949 27.2 dB SSIM 0.793 Try in SpecLab →	0.600	27.2	0.793	✓ Certified	Wiener 1949
9	Richardson-Lucy Richardson-Lucy Richardson 1972 24.5 dB SSIM 0.744 Try in SpecLab →	0.530	24.5	0.744	✓ Certified	Richardson 1972

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
🥇	DiffusionDF + gradient DiffusionDF + gradient Luo et al., arXiv 2023 (DF) Score 0.784 Correct & Reconstruct →	0.784	0.850 38.31 dB / 0.983	0.760 32.24 dB / 0.945	0.743 31.01 dB / 0.931	✓ Certified	Luo et al., arXiv 2023 (DF)
🥈	Restormer-DF + gradient Restormer-DF + gradient Zamir et al., CVPR 2022 (DF) Score 0.776 Correct & Reconstruct →	0.776	0.853 37.48 dB / 0.980	0.770 32.94 dB / 0.952	0.706 28.99 dB / 0.900	✓ Certified	Zamir et al., CVPR 2022 (DF)
🥉	SwinIR-DF + gradient SwinIR-DF + gradient Liang et al., ICCV 2021 (DF) Score 0.772 Correct & Reconstruct →	0.772	0.840 36.58 dB / 0.976	0.760 30.97 dB / 0.930	0.716 28.53 dB / 0.891	✓ Certified	Liang et al., ICCV 2021 (DF)
4	BM3D-DF + gradient BM3D-DF + gradient Dabov et al., IEEE TIP 2007 (DF adapt.) Score 0.738 Correct & Reconstruct →	0.738	0.768 30.87 dB / 0.929	0.739 30.27 dB / 0.921	0.707 28.83 dB / 0.897	✓ Certified	Dabov et al., IEEE TIP 2007 (DF adapt.)
5	Noise2Void-DF + gradient Noise2Void-DF + gradient Krull et al., CVPR 2019 (DF) Score 0.671 Correct & Reconstruct →	0.671	0.787 32.28 dB / 0.946	0.650 25.58 dB / 0.820	0.575 22.12 dB / 0.695	✓ Certified	Krull et al., CVPR 2019 (DF)
6	CARE-DF + gradient CARE-DF + gradient Weigert et al., Nat. Methods 2018 (DF) Score 0.655 Correct & Reconstruct →	0.655	0.787 33.29 dB / 0.955	0.639 25.2 dB / 0.808	0.540 21.8 dB / 0.681	✓ Certified	Weigert et al., Nat. Methods 2018 (DF)
7	Wiener-DF + gradient Wiener-DF + gradient Wiener, 1949 (DF adapt.) Score 0.630 Correct & Reconstruct →	0.630	0.651 25.35 dB / 0.813	0.640 24.62 dB / 0.790	0.600 24.02 dB / 0.769	✓ Certified	Wiener, 1949 (DF adapt.)
8	Richardson-Lucy + gradient Richardson-Lucy + gradient Richardson, JOSA 1972; Lucy, AJ 1974 Score 0.575 Correct & Reconstruct →	0.575	0.606 22.74 dB / 0.721	0.568 21.94 dB / 0.687	0.552 21.41 dB / 0.664	✓ Certified	Richardson, JOSA 1972; Lucy, AJ 1974
9	TV-DF + gradient TV-DF + gradient Rudin et al., Physica D 1992 (DF) Score 0.542 Correct & Reconstruct →	0.542	0.704 27.88 dB / 0.878	0.513 20.17 dB / 0.607	0.408 16.47 dB / 0.424	✓ Certified	Rudin et al., Physica D 1992 (DF)

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	Restormer-DF + gradient	0.853	37.48	0.98
2	DiffusionDF + gradient	0.850	38.31	0.983
3	SwinIR-DF + gradient	0.840	36.58	0.976
4	Noise2Void-DF + gradient	0.787	32.28	0.946
5	CARE-DF + gradient	0.787	33.29	0.955
6	BM3D-DF + gradient	0.768	30.87	0.929
7	TV-DF + gradient	0.704	27.88	0.878
8	Wiener-DF + gradient	0.651	25.35	0.813
9	Richardson-Lucy + gradient	0.606	22.74	0.721

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
condenser_na_vs_objective_na_ratio	1.14	1.32	-
stray_light	-1.0	2.0	relative
scattering_angle_range	-0.15	0.15	-

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	Restormer-DF + gradient	0.770	32.94	0.952
2	DiffusionDF + gradient	0.760	32.24	0.945
3	SwinIR-DF + gradient	0.760	30.97	0.93
4	BM3D-DF + gradient	0.739	30.27	0.921
5	Noise2Void-DF + gradient	0.650	25.58	0.82
6	Wiener-DF + gradient	0.640	24.62	0.79
7	CARE-DF + gradient	0.639	25.2	0.808
8	Richardson-Lucy + gradient	0.568	21.94	0.687
9	TV-DF + gradient	0.513	20.17	0.607

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
condenser_na_vs_objective_na_ratio	1.128	1.308	-
stray_light	-1.2	1.8	relative
scattering_angle_range	-0.15	0.15	-

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	DiffusionDF + gradient	0.743	31.01	0.931
2	SwinIR-DF + gradient	0.716	28.53	0.891
3	BM3D-DF + gradient	0.707	28.83	0.897
4	Restormer-DF + gradient	0.706	28.99	0.9
5	Wiener-DF + gradient	0.600	24.02	0.769
6	Noise2Void-DF + gradient	0.575	22.12	0.695
7	Richardson-Lucy + gradient	0.552	21.41	0.664
8	CARE-DF + gradient	0.540	21.8	0.681
9	TV-DF + gradient	0.408	16.47	0.424

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
condenser_na_vs_objective_na_ratio	1.158	1.338	-
stray_light	-0.7	2.3	relative
scattering_angle_range	-0.15	0.15	-

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
c_n	condenser_na_vs_objective_na_ratio	Condenser NA vs objective NA ratio (-)	1.2	1.26
s_l	stray_light	Stray light (relative)	0.0	1.0
s_a	scattering_angle_range	Scattering angle range (-)	0.0	0.0

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.