Polarimetric SAR (PolSAR)

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
🥇	DiffusionSAR DiffusionSAR Wei et al., NeurIPS 2024 35.42 dB SSIM 0.955 Checkpoint unavailable	0.818	35.42	0.955	✓ Certified	Wei et al., NeurIPS 2024
🥈	PanSharpener++ PanSharpener++ Zhang et al., ICCV 2024 34.58 dB SSIM 0.945 Checkpoint unavailable	0.799	34.58	0.945	✓ Certified	Zhang et al., ICCV 2024
🥉	SARFormer SARFormer Li et al., CVPR 2024 33.85 dB SSIM 0.932 Checkpoint unavailable	0.780	33.85	0.932	✓ Certified	Li et al., CVPR 2024
4	ScoreSAR ScoreSAR Johnson et al., ECCV 2025 32.44 dB SSIM 0.947 Checkpoint unavailable	0.764	32.44	0.947	✓ Certified	Johnson et al., ECCV 2025
5	SARDenoiserViT SARDenoiserViT Wang et al., ICCV 2024 32.31 dB SSIM 0.946 Checkpoint unavailable	0.762	32.31	0.946	✓ Certified	Wang et al., ICCV 2024
6	SAR-CAM SAR-CAM Cross-attention SAR, 2024 32.1 dB SSIM 0.912 Checkpoint unavailable	0.741	32.1	0.912	✓ Certified	Cross-attention SAR, 2024
7	SAR-ResNet SAR-ResNet Chen et al., IEEE TGRS 2022 30.88 dB SSIM 0.929 Checkpoint unavailable	0.729	30.88	0.929	✓ Certified	Chen et al., IEEE TGRS 2022
8	SAR-DRN SAR-DRN Zhang et al., RS 2018 30.6 dB SSIM 0.882 Checkpoint unavailable	0.701	30.6	0.882	✓ Certified	Zhang et al., RS 2018
9	SAR-BM3D SAR-BM3D Parrilli et al., IEEE TGRS 2012 27.2 dB SSIM 0.790 Try in SpecLab →	0.598	27.2	0.790	✓ Certified	Parrilli et al., IEEE TGRS 2012
10	Range-Doppler Range-Doppler SAR signal processing baseline 25.87 dB SSIM 0.828 Try in SpecLab →	0.595	25.87	0.828	✓ Certified	SAR signal processing baseline
11	Lee Filter Lee Filter Lee, IEEE TGRS 1980 25.62 dB SSIM 0.821 Try in SpecLab →	0.588	25.62	0.821	✓ Certified	Lee, IEEE TGRS 1980
12	Chirp Scaling Chirp Scaling Raney et al., IEEE TGRS 1994 24.82 dB SSIM 0.796 Try in SpecLab →	0.562	24.82	0.796	✓ Certified	Raney et al., IEEE TGRS 1994
13	Matched Filter Matched Filter Standard SAR focusing 23.5 dB SSIM 0.640 Try in SpecLab →	0.462	23.5	0.640	✓ Certified	Standard SAR focusing

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
🥇	SARFormer + gradient SARFormer + gradient Li et al., CVPR 2024 Score 0.746 Correct & Reconstruct →	0.746	0.792 32.59 dB / 0.949	0.742 29.8 dB / 0.914	0.703 27.83 dB / 0.877	✓ Certified	Li et al., CVPR 2024
🥈	DiffusionSAR + gradient DiffusionSAR + gradient Wei et al., NeurIPS 2024 Score 0.706 Correct & Reconstruct →	0.706	0.788 32.8 dB / 0.951	0.693 27.07 dB / 0.860	0.638 25.79 dB / 0.826	✓ Certified	Wei et al., NeurIPS 2024
🥉	PanSharpener++ + gradient PanSharpener++ + gradient Zhang et al., ICCV 2024 Score 0.702 Correct & Reconstruct →	0.702	0.776 31.9 dB / 0.942	0.677 26.59 dB / 0.848	0.653 25.47 dB / 0.817	✓ Certified	Zhang et al., ICCV 2024
4	SARDenoiserViT + gradient SARDenoiserViT + gradient Wang et al., ICCV 2024 Score 0.684 Correct & Reconstruct →	0.684	0.744 29.94 dB / 0.916	0.688 27.53 dB / 0.871	0.620 24.62 dB / 0.790	✓ Certified	Wang et al., ICCV 2024
5	ScoreSAR + gradient ScoreSAR + gradient Johnson et al., ECCV 2025 Score 0.682 Correct & Reconstruct →	0.682	0.773 31.43 dB / 0.936	0.643 24.9 dB / 0.799	0.630 24.67 dB / 0.792	✓ Certified	Johnson et al., ECCV 2025
6	SAR-CAM + gradient SAR-CAM + gradient Cross-attention SAR, 2024 Score 0.657 Correct & Reconstruct →	0.657	0.736 29.36 dB / 0.907	0.656 25.39 dB / 0.814	0.578 23.08 dB / 0.734	✓ Certified	Cross-attention SAR, 2024
7	SAR-DRN + gradient SAR-DRN + gradient Zhang et al., RS 2018 Score 0.635 Correct & Reconstruct →	0.635	0.715 28.37 dB / 0.888	0.623 24.73 dB / 0.793	0.566 22.82 dB / 0.724	✓ Certified	Zhang et al., RS 2018
8	Range-Doppler + gradient Range-Doppler + gradient SAR signal processing baseline Score 0.605 Correct & Reconstruct →	0.605	0.621 23.97 dB / 0.767	0.606 23.96 dB / 0.767	0.588 23.46 dB / 0.749	✓ Certified	SAR signal processing baseline
9	SAR-BM3D + gradient SAR-BM3D + gradient Parrilli et al., IEEE TGRS 2012 Score 0.591 Correct & Reconstruct →	0.591	0.638 24.36 dB / 0.781	0.605 23.49 dB / 0.750	0.531 21.3 dB / 0.659	✓ Certified	Parrilli et al., IEEE TGRS 2012
10	Chirp Scaling + gradient Chirp Scaling + gradient Raney et al., IEEE TGRS 1994 Score 0.577 Correct & Reconstruct →	0.577	0.582 22.21 dB / 0.699	0.593 23.17 dB / 0.738	0.557 21.51 dB / 0.669	✓ Certified	Raney et al., IEEE TGRS 1994
11	Lee Filter + gradient Lee Filter + gradient Lee, IEEE TGRS 1980 Score 0.570 Correct & Reconstruct →	0.570	0.636 24.04 dB / 0.770	0.560 22.21 dB / 0.699	0.515 20.89 dB / 0.641	✓ Certified	Lee, IEEE TGRS 1980
12	SAR-ResNet + gradient SAR-ResNet + gradient Chen et al., IEEE TGRS 2022 Score 0.569 Correct & Reconstruct →	0.569	0.746 29.69 dB / 0.912	0.541 21.38 dB / 0.663	0.420 17.68 dB / 0.484	✓ Certified	Chen et al., IEEE TGRS 2022
13	Matched Filter + gradient Matched Filter + gradient Standard SAR focusing Score 0.510 Correct & Reconstruct →	0.510	0.580 21.86 dB / 0.684	0.493 19.72 dB / 0.585	0.458 18.67 dB / 0.533	✓ Certified	Standard SAR focusing

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	SARFormer + gradient	0.792	32.59	0.949
2	DiffusionSAR + gradient	0.788	32.8	0.951
3	PanSharpener++ + gradient	0.776	31.9	0.942
4	ScoreSAR + gradient	0.773	31.43	0.936
5	SAR-ResNet + gradient	0.746	29.69	0.912
6	SARDenoiserViT + gradient	0.744	29.94	0.916
7	SAR-CAM + gradient	0.736	29.36	0.907
8	SAR-DRN + gradient	0.715	28.37	0.888
9	SAR-BM3D + gradient	0.638	24.36	0.781
10	Lee Filter + gradient	0.636	24.04	0.77
11	Range-Doppler + gradient	0.621	23.97	0.767
12	Chirp Scaling + gradient	0.582	22.21	0.699
13	Matched Filter + gradient	0.580	21.86	0.684

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
cross_talk_between_polarizations	-10.0	5.0	dB
channel_imbalance	-0.2	0.4	dB
faraday_rotation	-1.0	2.0	deg

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	SARFormer + gradient	0.742	29.8	0.914
2	DiffusionSAR + gradient	0.693	27.07	0.86
3	SARDenoiserViT + gradient	0.688	27.53	0.871
4	PanSharpener++ + gradient	0.677	26.59	0.848
5	SAR-CAM + gradient	0.656	25.39	0.814
6	ScoreSAR + gradient	0.643	24.9	0.799
7	SAR-DRN + gradient	0.623	24.73	0.793
8	Range-Doppler + gradient	0.606	23.96	0.767
9	SAR-BM3D + gradient	0.605	23.49	0.75
10	Chirp Scaling + gradient	0.593	23.17	0.738
11	Lee Filter + gradient	0.560	22.21	0.699
12	SAR-ResNet + gradient	0.541	21.38	0.663
13	Matched Filter + gradient	0.493	19.72	0.585

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
cross_talk_between_polarizations	-9.0	6.0	dB
channel_imbalance	-0.24	0.36	dB
faraday_rotation	-1.2	1.8	deg

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	SARFormer + gradient	0.703	27.83	0.877
2	PanSharpener++ + gradient	0.653	25.47	0.817
3	DiffusionSAR + gradient	0.638	25.79	0.826
4	ScoreSAR + gradient	0.630	24.67	0.792
5	SARDenoiserViT + gradient	0.620	24.62	0.79
6	Range-Doppler + gradient	0.588	23.46	0.749
7	SAR-CAM + gradient	0.578	23.08	0.734
8	SAR-DRN + gradient	0.566	22.82	0.724
9	Chirp Scaling + gradient	0.557	21.51	0.669
10	SAR-BM3D + gradient	0.531	21.3	0.659
11	Lee Filter + gradient	0.515	20.89	0.641
12	Matched Filter + gradient	0.458	18.67	0.533
13	SAR-ResNet + gradient	0.420	17.68	0.484

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
cross_talk_between_polarizations	-11.5	3.5	dB
channel_imbalance	-0.14	0.46	dB
faraday_rotation	-0.7	2.3	deg

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

F → M → D

F Fourier

M Modulation

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
c_b	cross_talk_between_polarizations	Cross-talk between polarizations (dB)	-5.0
c_i	channel_imbalance	Channel imbalance (dB)	0.2
f_r	faraday_rotation	Faraday rotation (deg)	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.