Focused Ion Beam SEM (FIB-SEM)

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
🥇	DiffFIB DiffFIB Gao et al. 2024 39.9 dB SSIM 0.959 Checkpoint unavailable	0.894	39.9	0.959	✓ Certified	Gao et al. 2024
🥈	PhysFIB PhysFIB Chen et al. 2024 38.6 dB SSIM 0.949 Checkpoint unavailable	0.868	38.6	0.949	✓ Certified	Chen et al. 2024
🥉	SwinFIB SwinFIB Wang et al. 2023 37.5 dB SSIM 0.939 Checkpoint unavailable	0.845	37.5	0.939	✓ Certified	Wang et al. 2023
4	TransFIB TransFIB Li et al. 2022 36.1 dB SSIM 0.923 Checkpoint unavailable	0.813	36.1	0.923	✓ Certified	Li et al. 2022
5	N2V-FIB N2V-FIB Krull et al. 2019 33.8 dB SSIM 0.891 Checkpoint unavailable	0.759	33.8	0.891	✓ Certified	Krull et al. 2019
6	DnCNN-FIB DnCNN-FIB Buchholz et al. 2019 31.9 dB SSIM 0.862 Checkpoint unavailable	0.713	31.9	0.862	✓ Certified	Buchholz et al. 2019
7	TV-FIB TV-FIB Rudin et al. 1992 29.4 dB SSIM 0.825 Try in SpecLab →	0.652	29.4	0.825	✓ Certified	Rudin et al. 1992
8	NLM-FIB NLM-FIB Buades et al. 2005 27.1 dB SSIM 0.789 Try in SpecLab →	0.596	27.1	0.789	✓ Certified	Buades et al. 2005
9	BM3D-FIB BM3D-FIB Dabov et al. 2007 25.3 dB SSIM 0.755 Try in SpecLab →	0.549	25.3	0.755	✓ Certified	Dabov et al. 2007

Dataset: PWM Benchmark (9 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
🥇	PhysFIB + gradient PhysFIB + gradient Chen et al., Nat. Commun. 2024 Score 0.785 Correct & Reconstruct →	0.785	0.829 36.17 dB / 0.974	0.779 33.01 dB / 0.953	0.746 31.26 dB / 0.934	✓ Certified	Chen et al., Nat. Commun. 2024
🥈	DiffFIB + gradient DiffFIB + gradient Gao et al., NeurIPS 2024 Score 0.783 Correct & Reconstruct →	0.783	0.844 36.91 dB / 0.978	0.773 31.86 dB / 0.941	0.731 30.9 dB / 0.930	✓ Certified	Gao et al., NeurIPS 2024
🥉	SwinFIB + gradient SwinFIB + gradient Wang et al., Nat. Commun. 2023 Score 0.772 Correct & Reconstruct →	0.772	0.836 35.89 dB / 0.973	0.778 33.07 dB / 0.953	0.703 29.06 dB / 0.901	✓ Certified	Wang et al., Nat. Commun. 2023
4	TransFIB + gradient TransFIB + gradient Li et al., Nat. Methods 2022 Score 0.741 Correct & Reconstruct →	0.741	0.796 33.38 dB / 0.956	0.749 31.15 dB / 0.933	0.679 27.7 dB / 0.874	✓ Certified	Li et al., Nat. Methods 2022
5	DnCNN-FIB + gradient DnCNN-FIB + gradient Buchholz et al., Nat. Methods 2019 Score 0.650 Correct & Reconstruct →	0.650	0.735 29.4 dB / 0.907	0.635 24.41 dB / 0.783	0.579 22.63 dB / 0.716	✓ Certified	Buchholz et al., Nat. Methods 2019
6	N2V-FIB + gradient N2V-FIB + gradient Krull et al., NeurIPS 2019 Score 0.628 Correct & Reconstruct →	0.628	0.788 32.07 dB / 0.943	0.608 23.22 dB / 0.740	0.489 19.44 dB / 0.572	✓ Certified	Krull et al., NeurIPS 2019
7	BM3D-FIB + gradient BM3D-FIB + gradient Dabov et al., IEEE TIP 2007 Score 0.620 Correct & Reconstruct →	0.620	0.635 24.08 dB / 0.771	0.622 24.19 dB / 0.775	0.603 23.65 dB / 0.756	✓ Certified	Dabov et al., IEEE TIP 2007
8	NLM-FIB + gradient NLM-FIB + gradient Buades et al., CVPR 2005 Score 0.605 Correct & Reconstruct →	0.605	0.675 25.93 dB / 0.830	0.590 23.32 dB / 0.743	0.551 21.97 dB / 0.689	✓ Certified	Buades et al., CVPR 2005
9	TV-FIB + gradient TV-FIB + gradient Rudin et al., Physica D 1992 Score 0.599 Correct & Reconstruct →	0.599	0.689 26.78 dB / 0.853	0.576 22.31 dB / 0.703	0.533 20.95 dB / 0.643	✓ Certified	Rudin et al., Physica D 1992

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	DiffFIB + gradient	0.844	36.91	0.978
2	SwinFIB + gradient	0.836	35.89	0.973
3	PhysFIB + gradient	0.829	36.17	0.974
4	TransFIB + gradient	0.796	33.38	0.956
5	N2V-FIB + gradient	0.788	32.07	0.943
6	DnCNN-FIB + gradient	0.735	29.4	0.907
7	TV-FIB + gradient	0.689	26.78	0.853
8	NLM-FIB + gradient	0.675	25.93	0.83
9	BM3D-FIB + gradient	0.635	24.08	0.771

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
slice_thickness_variation	-3.0	6.0	-
curtaining_artifact	-0.06	0.12	relative
charging	-60.0	120.0	V
drift_between_slices	-1.0	2.0	nm

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	PhysFIB + gradient	0.779	33.01	0.953
2	SwinFIB + gradient	0.778	33.07	0.953
3	DiffFIB + gradient	0.773	31.86	0.941
4	TransFIB + gradient	0.749	31.15	0.933
5	DnCNN-FIB + gradient	0.635	24.41	0.783
6	BM3D-FIB + gradient	0.622	24.19	0.775
7	N2V-FIB + gradient	0.608	23.22	0.74
8	NLM-FIB + gradient	0.590	23.32	0.743
9	TV-FIB + gradient	0.576	22.31	0.703

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
slice_thickness_variation	-3.6	5.4	-
curtaining_artifact	-0.072	0.108	relative
charging	-72.0	108.0	V
drift_between_slices	-1.2	1.8	nm

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	PhysFIB + gradient	0.746	31.26	0.934
2	DiffFIB + gradient	0.731	30.9	0.93
3	SwinFIB + gradient	0.703	29.06	0.901
4	TransFIB + gradient	0.679	27.7	0.874
5	BM3D-FIB + gradient	0.603	23.65	0.756
6	DnCNN-FIB + gradient	0.579	22.63	0.716
7	NLM-FIB + gradient	0.551	21.97	0.689
8	TV-FIB + gradient	0.533	20.95	0.643
9	N2V-FIB + gradient	0.489	19.44	0.572

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
slice_thickness_variation	-2.1	6.9	-
curtaining_artifact	-0.042	0.138	relative
charging	-42.0	138.0	V
drift_between_slices	-0.7	2.3	nm

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

S → C → D

S Sampling

C Convolution

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
s_t	slice_thickness_variation	Slice thickness variation (-)	3.0
c_a	curtaining_artifact	Curtaining artifact (relative)	0.06
c	charging	Charging (V)	60.0
d_b	drift_between_slices	Drift between slices (nm)	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.