Multispectral Satellite Imaging

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
🥇	FlowCompute FlowCompute Huang et al., ECCV 2025 38.35 dB SSIM 0.980 Try in SpecLab →	0.879	38.35	0.980	✓ Certified	Huang et al., ECCV 2025
🥈	DiffusionCompute DiffusionCompute Zhang et al., NeurIPS 2024 37.95 dB SSIM 0.978 Checkpoint unavailable	0.872	37.95	0.978	✓ Certified	Zhang et al., NeurIPS 2024
🥉	CompFormer CompFormer Liu et al., ICCV 2024 37.15 dB SSIM 0.972 Checkpoint unavailable	0.855	37.15	0.972	✓ Certified	Liu et al., ICCV 2024
4	Restormer Restormer Zamir et al., CVPR 2022 36.28 dB SSIM 0.968 Checkpoint unavailable	0.839	36.28	0.968	✓ Certified	Zamir et al., CVPR 2022
5	NAFNet NAFNet Chen et al., ICCV 2023 35.75 dB SSIM 0.962 Checkpoint unavailable	0.827	35.75	0.962	✓ Certified	Chen et al., ICCV 2023
6	SwinIR SwinIR Liang et al., ICCVW 2021 35.1 dB SSIM 0.955 Checkpoint unavailable	0.812	35.1	0.955	✓ Certified	Liang et al., ICCVW 2021
7	Deep Image Prior Deep Image Prior Ulyanov et al., CVPR 2018 33.72 dB SSIM 0.932 Checkpoint unavailable	0.778	33.72	0.932	✓ Certified	Ulyanov et al., CVPR 2018
8	PnP-ADMM PnP-ADMM ADMM + denoiser prior 30.85 dB SSIM 0.880 Try in SpecLab →	0.704	30.85	0.880	✓ Certified	ADMM + denoiser prior
9	Plug-and-Play Plug-and-Play Sreehari et al., IEEE TIP 2016 29.11 dB SSIM 0.902 Checkpoint unavailable	0.686	29.11	0.902	✓ Certified	Sreehari et al., IEEE TIP 2016
10	PnP-RED PnP-RED Romano et al., IEEE TIP 2017 30.18 dB SSIM 0.865 Try in SpecLab →	0.685	30.18	0.865	✓ Certified	Romano et al., IEEE TIP 2017
11	ART ART Gordon et al., 1970 28.2 dB SSIM 0.800 Try in SpecLab →	0.620	28.2	0.800	✓ Certified	Gordon et al., 1970
12	LSQR LSQR Paige & Saunders, 1982 27.8 dB SSIM 0.785 Try in SpecLab →	0.606	27.8	0.785	✓ Certified	Paige & Saunders, 1982
13	Tikhonov Tikhonov Tikhonov, 1963 26.5 dB SSIM 0.740 Try in SpecLab →	0.562	26.5	0.740	✓ Certified	Tikhonov, 1963

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
🥇	FlowCompute + gradient FlowCompute + gradient Huang et al., ECCV 2025 Score 0.765 Correct & Reconstruct →	0.765	0.846 36.73 dB / 0.977	0.743 30.49 dB / 0.924	0.706 28.54 dB / 0.892	✓ Certified	Huang et al., ECCV 2025
🥈	SwinIR + gradient SwinIR + gradient Liang et al., ICCVW 2021 Score 0.756 Correct & Reconstruct →	0.756	0.809 34.06 dB / 0.961	0.742 30.2 dB / 0.920	0.717 29.25 dB / 0.905	✓ Certified	Liang et al., ICCVW 2021
🥉	Restormer + gradient Restormer + gradient Zamir et al., CVPR 2022 Score 0.740 Correct & Reconstruct →	0.740	0.820 34.64 dB / 0.965	0.741 30.51 dB / 0.924	0.659 26.3 dB / 0.840	✓ Certified	Zamir et al., CVPR 2022
4	CompFormer + gradient CompFormer + gradient Liu et al., ICCV 2024 Score 0.732 Correct & Reconstruct →	0.732	0.811 34.85 dB / 0.967	0.720 29.79 dB / 0.914	0.664 26.46 dB / 0.844	✓ Certified	Liu et al., ICCV 2024
5	DiffusionCompute + gradient DiffusionCompute + gradient Zhang et al., NeurIPS 2024 Score 0.723 Correct & Reconstruct →	0.723	0.820 35.05 dB / 0.968	0.714 28.9 dB / 0.898	0.636 25.59 dB / 0.820	✓ Certified	Zhang et al., NeurIPS 2024
6	NAFNet + gradient NAFNet + gradient Chen et al., ICCV 2023 Score 0.719 Correct & Reconstruct →	0.719	0.816 34.27 dB / 0.963	0.715 28.36 dB / 0.888	0.627 25.09 dB / 0.805	✓ Certified	Chen et al., ICCV 2023
7	Deep Image Prior + gradient Deep Image Prior + gradient Ulyanov et al., CVPR 2018 Score 0.668 Correct & Reconstruct →	0.668	0.764 31.27 dB / 0.934	0.644 25.31 dB / 0.812	0.595 23.5 dB / 0.750	✓ Certified	Ulyanov et al., CVPR 2018
8	PnP-RED + gradient PnP-RED + gradient Romano et al., IEEE TIP 2017 Score 0.648 Correct & Reconstruct →	0.648	0.705 27.6 dB / 0.872	0.645 25.37 dB / 0.814	0.595 23.48 dB / 0.750	✓ Certified	Romano et al., IEEE TIP 2017
9	ART + gradient ART + gradient Gordon et al., J. Theor. Biol. 1970 Score 0.645 Correct & Reconstruct →	0.645	0.671 26.21 dB / 0.838	0.636 25.41 dB / 0.815	0.627 25.01 dB / 0.803	✓ Certified	Gordon et al., J. Theor. Biol. 1970
10	LSQR + gradient LSQR + gradient Paige & Saunders, TOMS 1982 Score 0.643 Correct & Reconstruct →	0.643	0.685 26.18 dB / 0.837	0.631 24.58 dB / 0.789	0.614 24.79 dB / 0.795	✓ Certified	Paige & Saunders, TOMS 1982
11	PnP-ADMM + gradient PnP-ADMM + gradient Venkatakrishnan et al., 2013 Score 0.620 Correct & Reconstruct →	0.620	0.717 28.24 dB / 0.886	0.607 24.08 dB / 0.771	0.537 21.68 dB / 0.676	✓ Certified	Venkatakrishnan et al., 2013
12	Tikhonov + gradient Tikhonov + gradient Tikhonov, Doklady Akad. Nauk SSSR 1963 Score 0.605 Correct & Reconstruct →	0.605	0.622 23.79 dB / 0.761	0.604 23.92 dB / 0.766	0.590 23.46 dB / 0.749	✓ Certified	Tikhonov, Doklady Akad. Nauk SSSR 1963
13	Plug-and-Play + gradient Plug-and-Play + gradient Sreehari et al., IEEE TIP 2016 Score 0.545 Correct & Reconstruct →	0.545	0.691 27.27 dB / 0.865	0.524 20.77 dB / 0.635	0.421 17.7 dB / 0.485	✓ Certified	Sreehari et al., IEEE TIP 2016

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	FlowCompute + gradient	0.846	36.73	0.977
2	Restormer + gradient	0.820	34.64	0.965
3	DiffusionCompute + gradient	0.820	35.05	0.968
4	NAFNet + gradient	0.816	34.27	0.963
5	CompFormer + gradient	0.811	34.85	0.967
6	SwinIR + gradient	0.809	34.06	0.961
7	Deep Image Prior + gradient	0.764	31.27	0.934
8	PnP-ADMM + gradient	0.717	28.24	0.886
9	PnP-RED + gradient	0.705	27.6	0.872
10	Plug-and-Play + gradient	0.691	27.27	0.865
11	LSQR + gradient	0.685	26.18	0.837
12	ART + gradient	0.671	26.21	0.838
13	Tikhonov + gradient	0.622	23.79	0.761

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
band_registration_error	-0.2	0.4	px
atmospheric_transmittance	0.83	0.89	-
radiometric_calibration	0.99	1.02	-
pointing_jitter	-0.1	0.2	px

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	FlowCompute + gradient	0.743	30.49	0.924
2	SwinIR + gradient	0.742	30.2	0.92
3	Restormer + gradient	0.741	30.51	0.924
4	CompFormer + gradient	0.720	29.79	0.914
5	NAFNet + gradient	0.715	28.36	0.888
6	DiffusionCompute + gradient	0.714	28.9	0.898
7	PnP-RED + gradient	0.645	25.37	0.814
8	Deep Image Prior + gradient	0.644	25.31	0.812
9	ART + gradient	0.636	25.41	0.815
10	LSQR + gradient	0.631	24.58	0.789
11	PnP-ADMM + gradient	0.607	24.08	0.771
12	Tikhonov + gradient	0.604	23.92	0.766
13	Plug-and-Play + gradient	0.524	20.77	0.635

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
band_registration_error	-0.24	0.36	px
atmospheric_transmittance	0.826	0.886	-
radiometric_calibration	0.988	1.018	-
pointing_jitter	-0.12	0.18	px

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	SwinIR + gradient	0.717	29.25	0.905
2	FlowCompute + gradient	0.706	28.54	0.892
3	CompFormer + gradient	0.664	26.46	0.844
4	Restormer + gradient	0.659	26.3	0.84
5	DiffusionCompute + gradient	0.636	25.59	0.82
6	NAFNet + gradient	0.627	25.09	0.805
7	ART + gradient	0.627	25.01	0.803
8	LSQR + gradient	0.614	24.79	0.795
9	Deep Image Prior + gradient	0.595	23.5	0.75
10	PnP-RED + gradient	0.595	23.48	0.75
11	Tikhonov + gradient	0.590	23.46	0.749
12	PnP-ADMM + gradient	0.537	21.68	0.676
13	Plug-and-Play + gradient	0.421	17.7	0.485

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
band_registration_error	-0.14	0.46	px
atmospheric_transmittance	0.836	0.896	-
radiometric_calibration	0.993	1.023	-
pointing_jitter	-0.07	0.23	px

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

M → Σ → D

M Modulation

Σ Summation

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
b_r	band_registration_error	Band registration error (px)	0.0	0.2
a_t	atmospheric_transmittance	Atmospheric transmittance (-)	0.85	0.87
r_c	radiometric_calibration	Radiometric calibration (-)	1.0	1.01
p_j	pointing_jitter	Pointing jitter (px)	0.0	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.