STED

STED Microscopy

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
🥇	ScoreMicro ScoreMicro Wei et al., ECCV 2025 38.48 dB SSIM 0.981 Checkpoint unavailable	0.882	38.48	0.981	✓ Certified	Wei et al., ECCV 2025
🥈	DiffDeconv DiffDeconv Huang et al., NeurIPS 2024 38.12 dB SSIM 0.979 Checkpoint unavailable	0.875	38.12	0.979	✓ Certified	Huang et al., NeurIPS 2024
🥉	Restormer+ Restormer+ Zamir et al., ICCV 2024 37.65 dB SSIM 0.975 Checkpoint unavailable	0.865	37.65	0.975	✓ Certified	Zamir et al., ICCV 2024
4	DeconvFormer DeconvFormer Chen et al., CVPR 2024 37.25 dB SSIM 0.972 Checkpoint unavailable	0.857	37.25	0.972	✓ Certified	Chen et al., CVPR 2024
5	ResUNet ResUNet DeCelle et al., Nat. Methods 2021 35.85 dB SSIM 0.964 Checkpoint unavailable	0.830	35.85	0.964	✓ Certified	DeCelle et al., Nat. Methods 2021
6	Restormer Restormer Zamir et al., CVPR 2022 35.8 dB SSIM 0.962 Checkpoint unavailable	0.828	35.8	0.962	✓ Certified	Zamir et al., CVPR 2022
7	U-Net U-Net Ronneberger et al., MICCAI 2015 35.15 dB SSIM 0.956 Checkpoint unavailable	0.814	35.15	0.956	✓ Certified	Ronneberger et al., MICCAI 2015
8	CARE CARE Weigert et al., Nat. Methods 2018 34.5 dB SSIM 0.948 Checkpoint unavailable	0.799	34.5	0.948	✓ Certified	Weigert et al., Nat. Methods 2018
9	PnP-DnCNN PnP-DnCNN Zhang et al., IEEE TIP 2017 31.2 dB SSIM 0.890 Try in SpecLab →	0.715	31.2	0.890	✓ Certified	Zhang et al., IEEE TIP 2017
10	PnP-FISTA PnP-FISTA Bai et al., 2020 30.42 dB SSIM 0.872 Try in SpecLab →	0.693	30.42	0.872	✓ Certified	Bai et al., 2020
11	TV-Deconvolution TV-Deconvolution TV-regularized deconvolution 29.5 dB SSIM 0.845 Try in SpecLab →	0.664	29.5	0.845	✓ Certified	TV-regularized deconvolution
12	Wiener Filter Wiener Filter Analytical baseline 28.35 dB SSIM 0.805 Try in SpecLab →	0.625	28.35	0.805	✓ Certified	Analytical baseline
13	Richardson-Lucy Richardson-Lucy Richardson 1972 / Lucy 1974 27.1 dB SSIM 0.770 Try in SpecLab →	0.587	27.1	0.770	✓ Certified	Richardson 1972 / Lucy 1974

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
🥇	Restormer+ + gradient Restormer+ + gradient Zamir et al., ICCV 2024 Score 0.785 Correct & Reconstruct →	0.785	0.816 34.8 dB / 0.966	0.789 33.15 dB / 0.954	0.751 31.97 dB / 0.942	✓ Certified	Zamir et al., ICCV 2024
🥈	DeconvFormer + gradient DeconvFormer + gradient Chen et al., CVPR 2024 Score 0.768 Correct & Reconstruct →	0.768	0.812 34.64 dB / 0.965	0.770 32.52 dB / 0.948	0.721 29.02 dB / 0.901	✓ Certified	Chen et al., CVPR 2024
🥉	ScoreMicro + gradient ScoreMicro + gradient Wei et al., ECCV 2025 Score 0.760 Correct & Reconstruct →	0.760	0.827 35.99 dB / 0.973	0.741 30.16 dB / 0.919	0.712 29.55 dB / 0.910	✓ Certified	Wei et al., ECCV 2025
4	Restormer + gradient Restormer + gradient Zamir et al., CVPR 2022 Score 0.744 Correct & Reconstruct →	0.744	0.795 33.73 dB / 0.959	0.738 30.58 dB / 0.925	0.700 27.67 dB / 0.874	✓ Certified	Zamir et al., CVPR 2022
5	DiffDeconv + gradient DiffDeconv + gradient Huang et al., NeurIPS 2024 Score 0.741 Correct & Reconstruct →	0.741	0.846 37.06 dB / 0.978	0.721 28.58 dB / 0.892	0.656 26.31 dB / 0.841	✓ Certified	Huang et al., NeurIPS 2024
6	ResUNet + gradient ResUNet + gradient DeCelle et al., Nat. Methods 2021 Score 0.723 Correct & Reconstruct →	0.723	0.816 34.28 dB / 0.963	0.696 28.3 dB / 0.887	0.657 25.45 dB / 0.816	✓ Certified	DeCelle et al., Nat. Methods 2021
7	PnP-DnCNN + gradient PnP-DnCNN + gradient Zhang et al., IEEE TIP 2017 Score 0.704 Correct & Reconstruct →	0.704	0.748 29.56 dB / 0.910	0.701 28.05 dB / 0.882	0.663 26.5 dB / 0.846	✓ Certified	Zhang et al., IEEE TIP 2017
8	U-Net + gradient U-Net + gradient Ronneberger et al., MICCAI 2015 Score 0.685 Correct & Reconstruct →	0.685	0.808 33.99 dB / 0.961	0.649 25.26 dB / 0.810	0.599 23.91 dB / 0.765	✓ Certified	Ronneberger et al., MICCAI 2015
9	PnP-FISTA + gradient PnP-FISTA + gradient Bai et al., 2020 Score 0.676 Correct & Reconstruct →	0.676	0.711 28.02 dB / 0.881	0.690 27.47 dB / 0.869	0.627 24.42 dB / 0.783	✓ Certified	Bai et al., 2020
10	TV-Deconvolution + gradient TV-Deconvolution + gradient Rudin et al., Phys. A 1992 Score 0.671 Correct & Reconstruct →	0.671	0.718 27.77 dB / 0.876	0.656 25.58 dB / 0.820	0.640 25.7 dB / 0.823	✓ Certified	Rudin et al., Phys. A 1992
11	CARE + gradient CARE + gradient Weigert et al., Nat. Methods 2018 Score 0.666 Correct & Reconstruct →	0.666	0.776 31.9 dB / 0.942	0.643 24.82 dB / 0.796	0.579 22.49 dB / 0.711	✓ Certified	Weigert et al., Nat. Methods 2018
12	Wiener Filter + gradient Wiener Filter + gradient Analytical baseline Score 0.654 Correct & Reconstruct →	0.654	0.698 26.92 dB / 0.856	0.647 25.54 dB / 0.819	0.618 24.61 dB / 0.790	✓ Certified	Analytical baseline
13	Richardson-Lucy + gradient Richardson-Lucy + gradient Richardson, JOSA 1972 / Lucy, AJ 1974 Score 0.631 Correct & Reconstruct →	0.631	0.673 25.66 dB / 0.822	0.621 24.3 dB / 0.779	0.598 23.43 dB / 0.748	✓ Certified	Richardson, JOSA 1972 / 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	DiffDeconv + gradient	0.846	37.06	0.978
2	ScoreMicro + gradient	0.827	35.99	0.973
3	Restormer+ + gradient	0.816	34.8	0.966
4	ResUNet + gradient	0.816	34.28	0.963
5	DeconvFormer + gradient	0.812	34.64	0.965
6	U-Net + gradient	0.808	33.99	0.961
7	Restormer + gradient	0.795	33.73	0.959
8	CARE + gradient	0.776	31.9	0.942
9	PnP-DnCNN + gradient	0.748	29.56	0.91
10	TV-Deconvolution + gradient	0.718	27.77	0.876
11	PnP-FISTA + gradient	0.711	28.02	0.881
12	Wiener Filter + gradient	0.698	26.92	0.856
13	Richardson-Lucy + gradient	0.673	25.66	0.822

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
depletion_power	-10.0	20.0	%
donut_alignment	-10.0	20.0	nm
saturation_intensity	-8.0	16.0	%

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	Restormer+ + gradient	0.789	33.15	0.954
2	DeconvFormer + gradient	0.770	32.52	0.948
3	ScoreMicro + gradient	0.741	30.16	0.919
4	Restormer + gradient	0.738	30.58	0.925
5	DiffDeconv + gradient	0.721	28.58	0.892
6	PnP-DnCNN + gradient	0.701	28.05	0.882
7	ResUNet + gradient	0.696	28.3	0.887
8	PnP-FISTA + gradient	0.690	27.47	0.869
9	TV-Deconvolution + gradient	0.656	25.58	0.82
10	U-Net + gradient	0.649	25.26	0.81
11	Wiener Filter + gradient	0.647	25.54	0.819
12	CARE + gradient	0.643	24.82	0.796
13	Richardson-Lucy + gradient	0.621	24.3	0.779

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
depletion_power	-12.0	18.0	%
donut_alignment	-12.0	18.0	nm
saturation_intensity	-9.6	14.4	%

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	Restormer+ + gradient	0.751	31.97	0.942
2	DeconvFormer + gradient	0.721	29.02	0.901
3	ScoreMicro + gradient	0.712	29.55	0.91
4	Restormer + gradient	0.700	27.67	0.874
5	PnP-DnCNN + gradient	0.663	26.5	0.846
6	ResUNet + gradient	0.657	25.45	0.816
7	DiffDeconv + gradient	0.656	26.31	0.841
8	TV-Deconvolution + gradient	0.640	25.7	0.823
9	PnP-FISTA + gradient	0.627	24.42	0.783
10	Wiener Filter + gradient	0.618	24.61	0.79
11	U-Net + gradient	0.599	23.91	0.765
12	Richardson-Lucy + gradient	0.598	23.43	0.748
13	CARE + gradient	0.579	22.49	0.711

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
depletion_power	-7.0	23.0	%
donut_alignment	-7.0	23.0	nm
saturation_intensity	-5.6	18.4	%

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

About the Imaging Modality

Stimulated emission depletion (STED) microscopy breaks the diffraction limit by overlaying the excitation focus with a doughnut-shaped depletion beam that forces fluorophores at the periphery back to the ground state via stimulated emission, effectively shrinking the fluorescent spot to 50 nm or below. The effective PSF width scales as d ~ lambda/(2*NA*sqrt(1 + I/I_s)) where I is the depletion intensity and I_s is the saturation intensity. Primary challenges include high depletion laser power causing photobleaching, and the photon-limited signal from the confined volume.

Principle

Stimulated Emission Depletion microscopy breaks the diffraction limit by using a donut-shaped depletion beam to force fluorophores at the periphery of the excitation spot back to the ground state via stimulated emission. Only fluorophores at the very center of the donut emit spontaneously, shrinking the effective PSF to 30-70 nm lateral resolution depending on depletion power.

How to Build the System

Combine an excitation laser (e.g., 640 nm pulsed) with a co-aligned depletion laser (775 nm pulsed, ~1 ns) that passes through a vortex phase plate to create the donut. Use a high-NA objective (100x 1.4 NA oil). Time-gate detection (1-6 ns after excitation pulse) to reject depletion photon leakage. Single-photon counting detectors (APDs or hybrid PMTs) are essential. Align the donut null precisely at the excitation center.

Common Reconstruction Algorithms

Richardson-Lucy deconvolution with STED PSF
Wiener deconvolution with known STED PSF
Deep-learning restoration (content-aware STED denoising)
Linear unmixing for multi-color STED
Time-gated STED (g-STED) background subtraction

Common Mistakes

Misaligned donut null causing asymmetric PSF and resolution loss
Excessive depletion power causing photobleaching of organic dyes
Depletion laser leaking into fluorescence detection channel
Insufficient time-gating, recording stimulated emission as signal
Using fluorophores with poor STED compatibility (low stimulated-emission cross-section)

How to Avoid Mistakes

Regularly check and optimize donut alignment using gold nanoparticle scattering
Use STED-optimized dyes (ATTO647N, SiR, Abberior STAR) and minimize power
Install proper spectral filters and use time-gating to reject depletion photons
Apply 1-6 ns detection gate synchronized with the pulsed excitation
Choose fluorophores specifically designed for STED with high photostability

Forward-Model Mismatch Cases

The widefield fallback uses a diffraction-limited PSF (sigma=2.0, ~250 nm resolution), but STED achieves 30-70 nm resolution by shrinking the effective PSF with the depletion donut — the fallback is 4-8x wider
The STED effective PSF depends on depletion beam power (d_eff = d_confocal / sqrt(1 + I_STED/I_sat)), making it fundamentally different from any fixed Gaussian — the fallback cannot model power-dependent resolution

How to Correct the Mismatch

Use the STED operator with the effective PSF that accounts for depletion beam intensity: PSF_eff has FWHM = lambda/(2*NA*sqrt(1 + I/I_sat)), typically 30-70 nm
Include the donut-shaped depletion profile and saturation intensity in the forward model; deconvolution with the correct sub-diffraction STED PSF recovers true super-resolution information

Experimental Setup — Signal Chain

Experimental setup diagram for STED Microscopy

Reconstruction Gallery — 4 Scenes × 3 Scenarios

Method: CPU_baseline | Mismatch: nominal (nominal=True, perturbed=False)

I Ideal II Mismatch (uncorrected) III Oracle correction

I — Ideal

Ground Truth

Measurement

Reconstruction

25.00718620382979 dB SSIM 0.7227278910193443

II — Mismatch

Ground Truth

Measurement

Reconstruction

21.677461578905422 dB SSIM 0.6331458337597847

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

8.505667670753299 dB SSIM 0.002961417354736242

I — Ideal

Ground Truth

Measurement

Reconstruction

33.590412225790416 dB SSIM 0.8983215849218369

II — Mismatch

Ground Truth

Measurement

Reconstruction

28.55119082801035 dB SSIM 0.6801687575426102

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

8.555717958906008 dB SSIM 0.0009550678731917688

I — Ideal

Ground Truth

Measurement

Reconstruction

28.66484879490426 dB SSIM 0.3828067152444124

II — Mismatch

Ground Truth

Measurement

Reconstruction

27.038340514472054 dB SSIM 0.219755172023952

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

5.3012455741441835 dB SSIM 0.001320565820223278

I — Ideal

Ground Truth

Measurement

Reconstruction

28.367189747421307 dB SSIM 0.7405355777404309

II — Mismatch

Ground Truth

Measurement

Reconstruction

25.677302889972143 dB SSIM 0.5944104196734429

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

5.569808978569241 dB SSIM 0.0011749395104755586

Mean PSNR Across All Scenes

28.907409242986443 dB

I — Ideal

25.73607395283999 dB

II — Mismatch

6.9831100455931825 dB

III — Oracle

Degradation: -3.2 dB Recovery: +-18.8 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	25.00718620382979	0.7227278910193443	21.677461578905422	0.6331458337597847	8.505667670753299	0.002961417354736242
scene_01	33.590412225790416	0.8983215849218369	28.55119082801035	0.6801687575426102	8.555717958906008	0.0009550678731917688
scene_02	28.66484879490426	0.3828067152444124	27.038340514472054	0.219755172023952	5.3012455741441835	0.001320565820223278
scene_03	28.367189747421307	0.7405355777404309	25.677302889972143	0.5944104196734429	5.569808978569241	0.0011749395104755586
Mean	28.907409242986443	0.6860979422315062	25.73607395283999	0.5318700457499475	6.9831100455931825	0.0016029976396567118

Experimental Setup

Instrument: Abberior STEDYCON / Leica TCS SP8 STED 3X

Objective: HC PL APO 100x / 1.40 NA oil STED WHITE

Pixel Size Nm: 20

Excitation Source: pulsed white-light laser (640 nm line)

Sted Depletion Nm: 775

Sted Laser: Onefive Katana HP (775 nm, 1.2 ns pulses)

Sted Power Mw: 200

Achieved Resolution Nm: 50

Dwell Time Us: 20

Detector: HyD hybrid detector (Leica) / APD

Dye: Abberior STAR RED / ATTO 647N

Key References

Hell & Wichmann, 'Breaking the diffraction resolution limit by stimulated emission', Optics Letters 19, 780-782 (1994)
Vicidomini et al., 'STED nanoscopy', Annual Review of Biophysics 47, 377-404 (2018)

Canonical Datasets

BioSR STED paired dataset (Zhang et al., Nature Methods 2023)
Abberior STED application note sample images

Spec DAG — Forward Model Pipeline

C(PSF_STED) → D(g, η₃)

C STED Effective PSF (PSF_STED)

D Avalanche Photodiode (g, η₃)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
ΔP	depletion_power	Depletion beam power error (%)	10.0
Δr	donut_alignment	Donut beam alignment error (nm)	10
ΔI_s	saturation_intensity	Saturation intensity error (%)	8.0

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.