Image Scanning Microscopy (ISM)

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 ScoreMicro Wei et al., ECCV 2025 38.48 dB SSIM 0.981 Checkpoint unavailable	0.882	38.48	0.981	✓ Certified	Wei et al., ECCV 2025
🥈	DiffDeconv DiffDeconv Huang et al., NeurIPS 2024 38.12 dB SSIM 0.979 Checkpoint unavailable	0.875	38.12	0.979	✓ Certified	Huang et al., NeurIPS 2024
🥉	Restormer+ Restormer+ Zamir et al., ICCV 2024 37.65 dB SSIM 0.975 Checkpoint unavailable	0.865	37.65	0.975	✓ Certified	Zamir et al., ICCV 2024
4	DeconvFormer DeconvFormer Chen et al., CVPR 2024 37.25 dB SSIM 0.972 Checkpoint unavailable	0.857	37.25	0.972	✓ Certified	Chen et al., CVPR 2024
5	ResUNet ResUNet DeCelle et al., Nat. Methods 2021 35.85 dB SSIM 0.964 Checkpoint unavailable	0.830	35.85	0.964	✓ Certified	DeCelle et al., Nat. Methods 2021
6	Restormer Restormer Zamir et al., CVPR 2022 35.8 dB SSIM 0.962 Checkpoint unavailable	0.828	35.8	0.962	✓ Certified	Zamir et al., CVPR 2022
7	U-Net U-Net Ronneberger et al., MICCAI 2015 35.15 dB SSIM 0.956 Checkpoint unavailable	0.814	35.15	0.956	✓ Certified	Ronneberger et al., MICCAI 2015
8	CARE CARE Weigert et al., Nat. Methods 2018 34.5 dB SSIM 0.948 Checkpoint unavailable	0.799	34.5	0.948	✓ Certified	Weigert et al., Nat. Methods 2018
9	PnP-DnCNN PnP-DnCNN Zhang et al., IEEE TIP 2017 31.2 dB SSIM 0.890 Try in SpecLab →	0.715	31.2	0.890	✓ Certified	Zhang et al., IEEE TIP 2017
10	PnP-FISTA PnP-FISTA Bai et al., 2020 30.42 dB SSIM 0.872 Try in SpecLab →	0.693	30.42	0.872	✓ Certified	Bai et al., 2020
11	TV-Deconvolution TV-Deconvolution TV-regularized deconvolution 29.5 dB SSIM 0.845 Try in SpecLab →	0.664	29.5	0.845	✓ Certified	TV-regularized deconvolution
12	Wiener Filter Wiener Filter Analytical baseline 28.35 dB SSIM 0.805 Try in SpecLab →	0.625	28.35	0.805	✓ Certified	Analytical baseline
13	Richardson-Lucy Richardson-Lucy Richardson 1972 / Lucy 1974 27.1 dB SSIM 0.770 Try in SpecLab →	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 Restormer+ + gradient Zamir et al., ICCV 2024 Score 0.785 Correct & Reconstruct →	0.785	0.839 36.16 dB / 0.974	0.767 31.89 dB / 0.941	0.750 31.35 dB / 0.935	✓ Certified	Zamir et al., ICCV 2024
🥈	DeconvFormer + gradient DeconvFormer + gradient Chen et al., CVPR 2024 Score 0.759 Correct & Reconstruct →	0.759	0.814 35.28 dB / 0.969	0.759 31.99 dB / 0.943	0.705 28.14 dB / 0.884	✓ Certified	Chen et al., CVPR 2024
🥉	ResUNet + gradient ResUNet + gradient DeCelle et al., Nat. Methods 2021 Score 0.739 Correct & Reconstruct →	0.739	0.817 34.67 dB / 0.966	0.719 28.34 dB / 0.888	0.680 27.35 dB / 0.866	✓ Certified	DeCelle et al., Nat. Methods 2021
4	ScoreMicro + gradient ScoreMicro + gradient Wei et al., ECCV 2025 Score 0.735 Correct & Reconstruct →	0.735	0.828 36.24 dB / 0.975	0.713 29.16 dB / 0.903	0.665 27.08 dB / 0.860	✓ Certified	Wei et al., ECCV 2025
5	Restormer + gradient Restormer + gradient Zamir et al., CVPR 2022 Score 0.731 Correct & Reconstruct →	0.731	0.795 33.72 dB / 0.959	0.748 30.85 dB / 0.929	0.651 25.55 dB / 0.819	✓ Certified	Zamir et al., CVPR 2022
6	DiffDeconv + gradient DiffDeconv + gradient Huang et al., NeurIPS 2024 Score 0.731 Correct & Reconstruct →	0.731	0.824 35.55 dB / 0.971	0.707 29.05 dB / 0.901	0.663 26.0 dB / 0.832	✓ Certified	Huang et al., NeurIPS 2024
7	CARE + gradient CARE + gradient Weigert et al., Nat. Methods 2018 Score 0.712 Correct & Reconstruct →	0.712	0.798 32.96 dB / 0.952	0.698 28.08 dB / 0.882	0.640 24.8 dB / 0.796	✓ Certified	Weigert et al., Nat. Methods 2018
8	TV-Deconvolution + gradient TV-Deconvolution + gradient Rudin et al., Phys. A 1992 Score 0.671 Correct & Reconstruct →	0.671	0.690 26.71 dB / 0.851	0.669 26.2 dB / 0.838	0.654 25.57 dB / 0.820	✓ Certified	Rudin et al., Phys. A 1992
9	PnP-DnCNN + gradient PnP-DnCNN + gradient Zhang et al., IEEE TIP 2017 Score 0.668 Correct & Reconstruct →	0.668	0.750 29.84 dB / 0.914	0.664 26.05 dB / 0.833	0.591 22.79 dB / 0.723	✓ Certified	Zhang et al., IEEE TIP 2017
10	U-Net + gradient U-Net + gradient Ronneberger et al., MICCAI 2015 Score 0.662 Correct & Reconstruct →	0.662	0.783 32.41 dB / 0.947	0.662 26.16 dB / 0.836	0.540 21.77 dB / 0.680	✓ Certified	Ronneberger et al., MICCAI 2015
11	Wiener Filter + gradient Wiener Filter + gradient Analytical baseline Score 0.635 Correct & Reconstruct →	0.635	0.662 25.4 dB / 0.815	0.636 24.68 dB / 0.792	0.606 23.7 dB / 0.758	✓ Certified	Analytical baseline
12	PnP-FISTA + gradient PnP-FISTA + gradient Bai et al., 2020 Score 0.627 Correct & Reconstruct →	0.627	0.707 27.74 dB / 0.875	0.622 24.53 dB / 0.787	0.551 21.47 dB / 0.667	✓ Certified	Bai et al., 2020
13	Richardson-Lucy + gradient Richardson-Lucy + gradient Richardson, JOSA 1972 / Lucy, AJ 1974 Score 0.619 Correct & Reconstruct →	0.619	0.637 24.4 dB / 0.782	0.626 24.51 dB / 0.786	0.594 23.46 dB / 0.749	✓ 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	Restormer+ + gradient	0.839	36.16	0.974
2	ScoreMicro + gradient	0.828	36.24	0.975
3	DiffDeconv + gradient	0.824	35.55	0.971
4	ResUNet + gradient	0.817	34.67	0.966
5	DeconvFormer + gradient	0.814	35.28	0.969
6	CARE + gradient	0.798	32.96	0.952
7	Restormer + gradient	0.795	33.72	0.959
8	U-Net + gradient	0.783	32.41	0.947
9	PnP-DnCNN + gradient	0.750	29.84	0.914
10	PnP-FISTA + gradient	0.707	27.74	0.875
11	TV-Deconvolution + gradient	0.690	26.71	0.851
12	Wiener Filter + gradient	0.662	25.4	0.815
13	Richardson-Lucy + gradient	0.637	24.4	0.782

Spec Ranges (2 parameters)

Parameter	Min	Max	Unit
detector_element_offset	-0.2	0.4	px
magnification_error	-1.0	2.0	relative

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+ + gradient	0.767	31.89	0.941
2	DeconvFormer + gradient	0.759	31.99	0.943
3	Restormer + gradient	0.748	30.85	0.929
4	ResUNet + gradient	0.719	28.34	0.888
5	ScoreMicro + gradient	0.713	29.16	0.903
6	DiffDeconv + gradient	0.707	29.05	0.901
7	CARE + gradient	0.698	28.08	0.882
8	TV-Deconvolution + gradient	0.669	26.2	0.838
9	PnP-DnCNN + gradient	0.664	26.05	0.833
10	U-Net + gradient	0.662	26.16	0.836
11	Wiener Filter + gradient	0.636	24.68	0.792
12	Richardson-Lucy + gradient	0.626	24.51	0.786
13	PnP-FISTA + gradient	0.622	24.53	0.787

Spec Ranges (2 parameters)

Parameter	Min	Max	Unit
detector_element_offset	-0.24	0.36	px
magnification_error	-1.2	1.8	relative

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	0.750	31.35	0.935
2	DeconvFormer + gradient	0.705	28.14	0.884
3	ResUNet + gradient	0.680	27.35	0.866
4	ScoreMicro + gradient	0.665	27.08	0.86
5	DiffDeconv + gradient	0.663	26.0	0.832
6	TV-Deconvolution + gradient	0.654	25.57	0.82
7	Restormer + gradient	0.651	25.55	0.819
8	CARE + gradient	0.640	24.8	0.796
9	Wiener Filter + gradient	0.606	23.7	0.758
10	Richardson-Lucy + gradient	0.594	23.46	0.749
11	PnP-DnCNN + gradient	0.591	22.79	0.723
12	PnP-FISTA + gradient	0.551	21.47	0.667
13	U-Net + gradient	0.540	21.77	0.68

Spec Ranges (2 parameters)

Parameter	Min	Max	Unit
detector_element_offset	-0.14	0.46	px
magnification_error	-0.7	2.3	relative

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
d_e	detector_element_offset	Detector element offset (px)	0.0	0.2
m_e	magnification_error	Magnification error (relative)	0.0	1.0

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.