High Dynamic Range (HDR) 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
🥇	DiffusionPhoto DiffusionPhoto Zhang et al., NeurIPS 2024 38.82 dB SSIM 0.978 Checkpoint unavailable	0.886	38.82	0.978	✓ Certified	Zhang et al., NeurIPS 2024
🥈	HDRFormer HDRFormer Eilertsen et al., ICCV 2024 38.15 dB SSIM 0.972 Checkpoint unavailable	0.872	38.15	0.972	✓ Certified	Eilertsen et al., ICCV 2024
🥉	DeblurGaussian DeblurGaussian Liang et al., CVPR 2024 37.68 dB SSIM 0.968 Checkpoint unavailable	0.862	37.68	0.968	✓ Certified	Liang et al., CVPR 2024
4	PhotoFormer PhotoFormer Zhang et al., ICCV 2024 36.48 dB SSIM 0.976 Checkpoint unavailable	0.846	36.48	0.976	✓ Certified	Zhang et al., ICCV 2024
5	Uformer Uformer Wang et al., CVPR 2022 36.2 dB SSIM 0.960 Checkpoint unavailable	0.833	36.2	0.960	✓ Certified	Wang et al., CVPR 2022
6	ScorePhoto ScorePhoto Wei et al., ECCV 2025 35.49 dB SSIM 0.971 Checkpoint unavailable	0.827	35.49	0.971	✓ Certified	Wei et al., ECCV 2025
7	U-Net U-Net Ronneberger et al., MICCAI 2015 35.1 dB SSIM 0.968 Checkpoint unavailable	0.819	35.1	0.968	✓ Certified	Ronneberger et al., MICCAI 2015
8	HDR-CNN HDR-CNN Eilertsen et al., ACM TOG 2017 34.9 dB SSIM 0.945 Checkpoint unavailable	0.804	34.9	0.945	✓ Certified	Eilertsen et al., ACM TOG 2017
9	LaplacianFormer LaplacianFormer Chen et al., CVPR 2022 33.16 dB SSIM 0.954 Checkpoint unavailable	0.780	33.16	0.954	✓ Certified	Chen et al., CVPR 2022
10	PnP-ADMM PnP-ADMM Venkatakrishnan et al., 2013 31.89 dB SSIM 0.941 Try in SpecLab →	0.752	31.89	0.941	✓ Certified	Venkatakrishnan et al., 2013
11	PnP-FFDNet PnP-FFDNet Zhang et al., 2017 31.45 dB SSIM 0.885 Try in SpecLab →	0.717	31.45	0.885	✓ Certified	Zhang et al., 2017
12	Lucy-Richardson Lucy-Richardson Lucy, AJ 1974 26.61 dB SSIM 0.848 Try in SpecLab →	0.617	26.61	0.848	✓ Certified	Lucy, AJ 1974
13	Laplacian Pyramid Laplacian Pyramid Burt & Adelson, TPAMI 1983 26.42 dB SSIM 0.843 Try in SpecLab →	0.612	26.42	0.843	✓ Certified	Burt & Adelson, TPAMI 1983
14	Wiener-Deconv Wiener-Deconv Analytical baseline 27.8 dB SSIM 0.780 Try in SpecLab →	0.603	27.8	0.780	✓ Certified	Analytical baseline

Dataset: PWM Benchmark (14 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
🥇	DeblurGaussian + gradient DeblurGaussian + gradient Liang et al., CVPR 2024 Score 0.791 Correct & Reconstruct →	0.791	0.818 35.1 dB / 0.968	0.787 33.18 dB / 0.954	0.767 32.69 dB / 0.950	✓ Certified	Liang et al., CVPR 2024
🥈	HDRFormer + gradient HDRFormer + gradient Eilertsen et al., ICCV 2024 Score 0.781 Correct & Reconstruct →	0.781	0.822 35.22 dB / 0.969	0.772 32.07 dB / 0.943	0.748 31.21 dB / 0.934	✓ Certified	Eilertsen et al., ICCV 2024
🥉	PhotoFormer + gradient PhotoFormer + gradient Zhang et al., ICCV 2024 Score 0.770 Correct & Reconstruct →	0.770	0.802 33.67 dB / 0.958	0.779 32.25 dB / 0.945	0.730 30.7 dB / 0.927	✓ Certified	Zhang et al., ICCV 2024
4	DiffusionPhoto + gradient DiffusionPhoto + gradient Zhang et al., NeurIPS 2024 Score 0.739 Correct & Reconstruct →	0.739	0.853 37.47 dB / 0.980	0.709 28.8 dB / 0.897	0.654 25.4 dB / 0.815	✓ Certified	Zhang et al., NeurIPS 2024
5	Uformer + gradient Uformer + gradient Wang et al., CVPR 2022 Score 0.729 Correct & Reconstruct →	0.729	0.822 34.95 dB / 0.967	0.730 30.06 dB / 0.918	0.636 24.87 dB / 0.798	✓ Certified	Wang et al., CVPR 2022
6	HDR-CNN + gradient HDR-CNN + gradient Eilertsen et al., ACM TOG 2017 Score 0.727 Correct & Reconstruct →	0.727	0.804 33.38 dB / 0.956	0.713 29.13 dB / 0.903	0.663 26.22 dB / 0.838	✓ Certified	Eilertsen et al., ACM TOG 2017
7	U-Net + gradient U-Net + gradient Ronneberger et al., MICCAI 2015 Score 0.720 Correct & Reconstruct →	0.720	0.809 34.03 dB / 0.961	0.694 27.35 dB / 0.866	0.656 25.48 dB / 0.817	✓ Certified	Ronneberger et al., MICCAI 2015
8	ScorePhoto + gradient ScorePhoto + gradient Wei et al., ECCV 2025 Score 0.719 Correct & Reconstruct →	0.719	0.792 33.34 dB / 0.956	0.701 28.59 dB / 0.893	0.664 26.74 dB / 0.852	✓ Certified	Wei et al., ECCV 2025
9	LaplacianFormer + gradient LaplacianFormer + gradient Chen et al., CVPR 2022 Score 0.672 Correct & Reconstruct →	0.672	0.755 30.49 dB / 0.924	0.658 25.33 dB / 0.812	0.604 24.01 dB / 0.769	✓ Certified	Chen et al., CVPR 2022
10	PnP-ADMM + gradient PnP-ADMM + gradient Venkatakrishnan et al., 2013 Score 0.670 Correct & Reconstruct →	0.670	0.732 29.0 dB / 0.900	0.674 26.84 dB / 0.854	0.604 23.63 dB / 0.755	✓ Certified	Venkatakrishnan et al., 2013
11	PnP-FFDNet + gradient PnP-FFDNet + gradient Zhang et al., 2017 Score 0.645 Correct & Reconstruct →	0.645	0.728 28.86 dB / 0.898	0.636 24.79 dB / 0.795	0.572 22.21 dB / 0.699	✓ Certified	Zhang et al., 2017
12	Wiener-Deconv + gradient Wiener-Deconv + gradient Analytical baseline Score 0.642 Correct & Reconstruct →	0.642	0.691 26.7 dB / 0.851	0.634 25.06 dB / 0.804	0.601 23.33 dB / 0.744	✓ Certified	Analytical baseline
13	Laplacian Pyramid + gradient Laplacian Pyramid + gradient Burt & Adelson, TPAMI 1983 Score 0.619 Correct & Reconstruct →	0.619	0.624 23.94 dB / 0.766	0.637 24.9 dB / 0.799	0.595 23.03 dB / 0.732	✓ Certified	Burt & Adelson, TPAMI 1983
14	Lucy-Richardson + gradient Lucy-Richardson + gradient Lucy, AJ 1974 Score 0.596 Correct & Reconstruct →	0.596	0.631 24.33 dB / 0.780	0.612 24.03 dB / 0.770	0.545 22.05 dB / 0.692	✓ Certified	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	DiffusionPhoto + gradient	0.853	37.47	0.98
2	HDRFormer + gradient	0.822	35.22	0.969
3	Uformer + gradient	0.822	34.95	0.967
4	DeblurGaussian + gradient	0.818	35.1	0.968
5	U-Net + gradient	0.809	34.03	0.961
6	HDR-CNN + gradient	0.804	33.38	0.956
7	PhotoFormer + gradient	0.802	33.67	0.958
8	ScorePhoto + gradient	0.792	33.34	0.956
9	LaplacianFormer + gradient	0.755	30.49	0.924
10	PnP-ADMM + gradient	0.732	29.0	0.9
11	PnP-FFDNet + gradient	0.728	28.86	0.898
12	Wiener-Deconv + gradient	0.691	26.7	0.851
13	Lucy-Richardson + gradient	0.631	24.33	0.78
14	Laplacian Pyramid + gradient	0.624	23.94	0.766

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
camera_response_function_error	-2.0	4.0	-
exposure_ratio_error	-2.0	4.0	-
ghost_artifact_(motion_between_exposures)	-1.0	2.0	px

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	DeblurGaussian + gradient	0.787	33.18	0.954
2	PhotoFormer + gradient	0.779	32.25	0.945
3	HDRFormer + gradient	0.772	32.07	0.943
4	Uformer + gradient	0.730	30.06	0.918
5	HDR-CNN + gradient	0.713	29.13	0.903
6	DiffusionPhoto + gradient	0.709	28.8	0.897
7	ScorePhoto + gradient	0.701	28.59	0.893
8	U-Net + gradient	0.694	27.35	0.866
9	PnP-ADMM + gradient	0.674	26.84	0.854
10	LaplacianFormer + gradient	0.658	25.33	0.812
11	Laplacian Pyramid + gradient	0.637	24.9	0.799
12	PnP-FFDNet + gradient	0.636	24.79	0.795
13	Wiener-Deconv + gradient	0.634	25.06	0.804
14	Lucy-Richardson + gradient	0.612	24.03	0.77

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
camera_response_function_error	-2.4	3.6	-
exposure_ratio_error	-2.4	3.6	-
ghost_artifact_(motion_between_exposures)	-1.2	1.8	px

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	DeblurGaussian + gradient	0.767	32.69	0.95
2	HDRFormer + gradient	0.748	31.21	0.934
3	PhotoFormer + gradient	0.730	30.7	0.927
4	ScorePhoto + gradient	0.664	26.74	0.852
5	HDR-CNN + gradient	0.663	26.22	0.838
6	U-Net + gradient	0.656	25.48	0.817
7	DiffusionPhoto + gradient	0.654	25.4	0.815
8	Uformer + gradient	0.636	24.87	0.798
9	LaplacianFormer + gradient	0.604	24.01	0.769
10	PnP-ADMM + gradient	0.604	23.63	0.755
11	Wiener-Deconv + gradient	0.601	23.33	0.744
12	Laplacian Pyramid + gradient	0.595	23.03	0.732
13	PnP-FFDNet + gradient	0.572	22.21	0.699
14	Lucy-Richardson + gradient	0.545	22.05	0.692

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
camera_response_function_error	-1.4	4.6	-
exposure_ratio_error	-1.4	4.6	-
ghost_artifact_(motion_between_exposures)	-0.7	2.3	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	Perturbed
c_r	camera_response_function_error	Camera response function error (-)	2.0
e_r	exposure_ratio_error	Exposure ratio error (-)	2.0
g_a	ghost_artifact_(motion_between_exposures)	Ghost artifact (motion between exposures) (px)	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.