Magnetic Particle Imaging (MPI)

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
🥇	SwinIR SwinIR Liang et al., ICCVW 2021 34.1 dB SSIM 0.942 Checkpoint unavailable	0.789	34.1	0.942	✓ Certified	Liang et al., ICCVW 2021
🥈	ScoreExperimental ScoreExperimental Wei et al., 2025 33.54 dB SSIM 0.957 Checkpoint unavailable	0.787	33.54	0.957	✓ Certified	Wei et al., 2025
🥉	ExpFormer ExpFormer Experimental science transformer, 2024 32.23 dB SSIM 0.945 Checkpoint unavailable	0.760	32.23	0.945	✓ Certified	Experimental science transformer, 2024
4	Domain-Adapted-CNN Domain-Adapted-CNN Domain adaptation CNN 32.16 dB SSIM 0.944 Checkpoint unavailable	0.758	32.16	0.944	✓ Certified	Domain adaptation CNN
5	ResUNet ResUNet Residual U-Net baseline 32.6 dB SSIM 0.915 Checkpoint unavailable	0.751	32.6	0.915	✓ Certified	Residual U-Net baseline
6	DiffusionExperimental DiffusionExperimental Zhang et al., 2024 30.6 dB SSIM 0.926 Checkpoint unavailable	0.723	30.6	0.926	✓ Certified	Zhang et al., 2024
7	PnP-ADMM PnP-ADMM ADMM + denoiser prior 29.57 dB SSIM 0.910 Try in SpecLab →	0.698	29.57	0.910	✓ Certified	ADMM + denoiser prior
8	PnP-RED PnP-RED Romano et al., IEEE TIP 2017 28.9 dB SSIM 0.835 Try in SpecLab →	0.649	28.9	0.835	✓ Certified	Romano et al., IEEE TIP 2017
9	Wiener Filter Wiener Filter Wiener filtering baseline 27.39 dB SSIM 0.867 Try in SpecLab →	0.640	27.39	0.867	✓ Certified	Wiener filtering baseline
10	Matched Filter Matched Filter Optimal linear filter 24.65 dB SSIM 0.791 Try in SpecLab →	0.556	24.65	0.791	✓ Certified	Optimal linear filter
11	Tikhonov Tikhonov Analytical baseline 25.4 dB SSIM 0.710 Try in SpecLab →	0.528	25.4	0.710	✓ Certified	Analytical baseline

Dataset: PWM Benchmark (11 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
🥇	ScoreExperimental + gradient ScoreExperimental + gradient Wei et al., 2025 Score 0.691 Correct & Reconstruct →	0.691	0.765 31.57 dB / 0.938	0.676 26.95 dB / 0.857	0.633 24.56 dB / 0.788	✓ Certified	Wei et al., 2025
🥈	SwinIR + gradient SwinIR + gradient Liang et al., ICCVW 2021 Score 0.687 Correct & Reconstruct →	0.687	0.793 32.39 dB / 0.947	0.691 27.0 dB / 0.858	0.576 22.51 dB / 0.711	✓ Certified	Liang et al., ICCVW 2021
🥉	Domain-Adapted-CNN + gradient Domain-Adapted-CNN + gradient Domain adaptation CNN Score 0.671 Correct & Reconstruct →	0.671	0.737 29.25 dB / 0.905	0.653 25.1 dB / 0.805	0.624 24.52 dB / 0.787	✓ Certified	Domain adaptation CNN
4	ExpFormer + gradient ExpFormer + gradient Experimental science transformer, 2024 Score 0.648 Correct & Reconstruct →	0.648	0.766 30.55 dB / 0.925	0.638 24.63 dB / 0.790	0.541 21.67 dB / 0.676	✓ Certified	Experimental science transformer, 2024
5	DiffusionExperimental + gradient DiffusionExperimental + gradient Zhang et al., 2024 Score 0.647 Correct & Reconstruct →	0.647	0.715 28.38 dB / 0.889	0.641 25.18 dB / 0.808	0.586 23.24 dB / 0.740	✓ Certified	Zhang et al., 2024
6	ResUNet + gradient ResUNet + gradient Residual U-Net baseline Score 0.644 Correct & Reconstruct →	0.644	0.771 30.94 dB / 0.930	0.608 23.31 dB / 0.743	0.552 21.93 dB / 0.687	✓ Certified	Residual U-Net baseline
7	PnP-RED + gradient PnP-RED + gradient Romano et al., IEEE TIP 2017 Score 0.642 Correct & Reconstruct →	0.642	0.686 27.04 dB / 0.859	0.627 24.85 dB / 0.797	0.614 24.1 dB / 0.772	✓ Certified	Romano et al., IEEE TIP 2017
8	Wiener Filter + gradient Wiener Filter + gradient Wiener filtering baseline Score 0.618 Correct & Reconstruct →	0.618	0.675 25.64 dB / 0.822	0.627 24.49 dB / 0.785	0.552 21.43 dB / 0.665	✓ Certified	Wiener filtering baseline
9	PnP-ADMM + gradient PnP-ADMM + gradient ADMM + denoiser prior Score 0.612 Correct & Reconstruct →	0.612	0.694 27.02 dB / 0.859	0.600 23.87 dB / 0.764	0.541 21.81 dB / 0.682	✓ Certified	ADMM + denoiser prior
10	Tikhonov + gradient Tikhonov + gradient Tikhonov, Doklady 1963 Score 0.570 Correct & Reconstruct →	0.570	0.602 23.14 dB / 0.737	0.572 22.36 dB / 0.705	0.536 21.2 dB / 0.655	✓ Certified	Tikhonov, Doklady 1963
11	Matched Filter + gradient Matched Filter + gradient Optimal linear filter Score 0.565 Correct & Reconstruct →	0.565	0.620 23.52 dB / 0.751	0.574 22.35 dB / 0.705	0.501 20.41 dB / 0.618	✓ Certified	Optimal linear filter

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	SwinIR + gradient	0.793	32.39	0.947
2	ResUNet + gradient	0.771	30.94	0.93
3	ExpFormer + gradient	0.766	30.55	0.925
4	ScoreExperimental + gradient	0.765	31.57	0.938
5	Domain-Adapted-CNN + gradient	0.737	29.25	0.905
6	DiffusionExperimental + gradient	0.715	28.38	0.889
7	PnP-ADMM + gradient	0.694	27.02	0.859
8	PnP-RED + gradient	0.686	27.04	0.859
9	Wiener Filter + gradient	0.675	25.64	0.822
10	Matched Filter + gradient	0.620	23.52	0.751
11	Tikhonov + gradient	0.602	23.14	0.737

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
drive_field_amplitude	24.4	26.2	mT
selection_field_gradient	2.4	2.7	T/m
particle_relaxation_time	1.8	2.4	us
receive_coil_sensitivity	0.97	1.06	-

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	SwinIR + gradient	0.691	27.0	0.858
2	ScoreExperimental + gradient	0.676	26.95	0.857
3	Domain-Adapted-CNN + gradient	0.653	25.1	0.805
4	DiffusionExperimental + gradient	0.641	25.18	0.808
5	ExpFormer + gradient	0.638	24.63	0.79
6	PnP-RED + gradient	0.627	24.85	0.797
7	Wiener Filter + gradient	0.627	24.49	0.785
8	ResUNet + gradient	0.608	23.31	0.743
9	PnP-ADMM + gradient	0.600	23.87	0.764
10	Matched Filter + gradient	0.574	22.35	0.705
11	Tikhonov + gradient	0.572	22.36	0.705

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
drive_field_amplitude	24.28	26.08	mT
selection_field_gradient	2.38	2.68	T/m
particle_relaxation_time	1.76	2.36	us
receive_coil_sensitivity	0.964	1.054	-

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	ScoreExperimental + gradient	0.633	24.56	0.788
2	Domain-Adapted-CNN + gradient	0.624	24.52	0.787
3	PnP-RED + gradient	0.614	24.1	0.772
4	DiffusionExperimental + gradient	0.586	23.24	0.74
5	SwinIR + gradient	0.576	22.51	0.711
6	ResUNet + gradient	0.552	21.93	0.687
7	Wiener Filter + gradient	0.552	21.43	0.665
8	ExpFormer + gradient	0.541	21.67	0.676
9	PnP-ADMM + gradient	0.541	21.81	0.682
10	Tikhonov + gradient	0.536	21.2	0.655
11	Matched Filter + gradient	0.501	20.41	0.618

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
drive_field_amplitude	24.58	26.38	mT
selection_field_gradient	2.43	2.73	T/m
particle_relaxation_time	1.86	2.46	us
receive_coil_sensitivity	0.979	1.069	-

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 → F → D

M Modulation

F Fourier

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
d_f	drive_field_amplitude	Drive field amplitude (mT)	25.0	25.6
s_f	selection_field_gradient	Selection field gradient (T/m)	2.5	2.6
p_r	particle_relaxation_time	Particle relaxation time (us)	2.0	2.2
r_c	receive_coil_sensitivity	Receive coil sensitivity (-)	1.0	1.03

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.