Two-Photon

Two-Photon / Multiphoton 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 0.882 38.48 0.981 ✓ Certified Wei et al., ECCV 2025
🥈 DiffDeconv 0.875 38.12 0.979 ✓ Certified Huang et al., NeurIPS 2024
🥉 Restormer+ 0.865 37.65 0.975 ✓ Certified Zamir et al., ICCV 2024
4 DeconvFormer 0.857 37.25 0.972 ✓ Certified Chen et al., CVPR 2024
5 ResUNet 0.830 35.85 0.964 ✓ Certified DeCelle et al., Nat. Methods 2021
6 Restormer 0.828 35.8 0.962 ✓ Certified Zamir et al., CVPR 2022
7 U-Net 0.814 35.15 0.956 ✓ Certified Ronneberger et al., MICCAI 2015
8 CARE 0.799 34.5 0.948 ✓ Certified Weigert et al., Nat. Methods 2018
9 PnP-DnCNN 0.715 31.2 0.890 ✓ Certified Zhang et al., IEEE TIP 2017
10 PnP-FISTA 0.693 30.42 0.872 ✓ Certified Bai et al., 2020
11 TV-Deconvolution 0.664 29.5 0.845 ✓ Certified TV-regularized deconvolution
12 Wiener Filter 0.625 28.35 0.805 ✓ Certified Analytical baseline
13 Richardson-Lucy 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
🥇 DeconvFormer + gradient 0.796
0.835
35.79 dB / 0.972
0.794
33.38 dB / 0.956
0.760
31.27 dB / 0.934
✓ Certified Chen et al., CVPR 2024
🥈 Restormer+ + gradient 0.767
0.818
35.56 dB / 0.971
0.762
31.26 dB / 0.934
0.720
29.55 dB / 0.910
✓ Certified Zamir et al., ICCV 2024
🥉 Restormer + gradient 0.760
0.816
34.39 dB / 0.964
0.753
31.46 dB / 0.937
0.710
29.09 dB / 0.902
✓ Certified Zamir et al., CVPR 2022
4 DiffDeconv + gradient 0.749
0.846
37.08 dB / 0.978
0.726
29.97 dB / 0.916
0.675
26.67 dB / 0.850
✓ Certified Huang et al., NeurIPS 2024
5 ScoreMicro + gradient 0.747
0.849
37.12 dB / 0.979
0.719
28.5 dB / 0.891
0.673
26.99 dB / 0.858
✓ Certified Wei et al., ECCV 2025
6 U-Net + gradient 0.731
0.785
32.58 dB / 0.949
0.722
28.54 dB / 0.892
0.686
27.02 dB / 0.859
✓ Certified Ronneberger et al., MICCAI 2015
7 CARE + gradient 0.696
0.799
32.85 dB / 0.951
0.660
26.45 dB / 0.844
0.628
24.26 dB / 0.778
✓ Certified Weigert et al., Nat. Methods 2018
8 ResUNet + gradient 0.681
0.797
34.07 dB / 0.961
0.662
25.79 dB / 0.826
0.583
23.42 dB / 0.747
✓ Certified DeCelle et al., Nat. Methods 2021
9 TV-Deconvolution + gradient 0.664
0.693
27.21 dB / 0.863
0.680
27.02 dB / 0.859
0.620
25.05 dB / 0.804
✓ Certified Rudin et al., Phys. A 1992
10 PnP-DnCNN + gradient 0.660
0.749
29.64 dB / 0.911
0.630
24.77 dB / 0.795
0.600
23.52 dB / 0.751
✓ Certified Zhang et al., IEEE TIP 2017
11 Wiener Filter + gradient 0.654
0.701
27.17 dB / 0.862
0.656
25.95 dB / 0.831
0.604
23.52 dB / 0.751
✓ Certified Analytical baseline
12 Richardson-Lucy + gradient 0.637
0.674
25.8 dB / 0.826
0.631
24.31 dB / 0.779
0.606
24.36 dB / 0.781
✓ Certified Richardson, JOSA 1972 / Lucy, AJ 1974
13 PnP-FISTA + gradient 0.609
0.709
27.9 dB / 0.879
0.599
23.18 dB / 0.738
0.519
20.34 dB / 0.615
✓ Certified Bai et al., 2020

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 ScoreMicro + gradient 0.849 37.12 0.979
2 DiffDeconv + gradient 0.846 37.08 0.978
3 DeconvFormer + gradient 0.835 35.79 0.972
4 Restormer+ + gradient 0.818 35.56 0.971
5 Restormer + gradient 0.816 34.39 0.964
6 CARE + gradient 0.799 32.85 0.951
7 ResUNet + gradient 0.797 34.07 0.961
8 U-Net + gradient 0.785 32.58 0.949
9 PnP-DnCNN + gradient 0.749 29.64 0.911
10 PnP-FISTA + gradient 0.709 27.9 0.879
11 Wiener Filter + gradient 0.701 27.17 0.862
12 TV-Deconvolution + gradient 0.693 27.21 0.863
13 Richardson-Lucy + gradient 0.674 25.8 0.826
Spec Ranges (3 parameters)
Parameter Min Max Unit
pulse_width 60.0 180.0 fs
gdd -500.0 1000.0 fs²
scattering -10.0 20.0 %
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 DeconvFormer + gradient 0.794 33.38 0.956
2 Restormer+ + gradient 0.762 31.26 0.934
3 Restormer + gradient 0.753 31.46 0.937
4 DiffDeconv + gradient 0.726 29.97 0.916
5 U-Net + gradient 0.722 28.54 0.892
6 ScoreMicro + gradient 0.719 28.5 0.891
7 TV-Deconvolution + gradient 0.680 27.02 0.859
8 ResUNet + gradient 0.662 25.79 0.826
9 CARE + gradient 0.660 26.45 0.844
10 Wiener Filter + gradient 0.656 25.95 0.831
11 Richardson-Lucy + gradient 0.631 24.31 0.779
12 PnP-DnCNN + gradient 0.630 24.77 0.795
13 PnP-FISTA + gradient 0.599 23.18 0.738
Spec Ranges (3 parameters)
Parameter Min Max Unit
pulse_width 52.0 172.0 fs
gdd -600.0 900.0 fs²
scattering -12.0 18.0 %
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 DeconvFormer + gradient 0.760 31.27 0.934
2 Restormer+ + gradient 0.720 29.55 0.91
3 Restormer + gradient 0.710 29.09 0.902
4 U-Net + gradient 0.686 27.02 0.859
5 DiffDeconv + gradient 0.675 26.67 0.85
6 ScoreMicro + gradient 0.673 26.99 0.858
7 CARE + gradient 0.628 24.26 0.778
8 TV-Deconvolution + gradient 0.620 25.05 0.804
9 Richardson-Lucy + gradient 0.606 24.36 0.781
10 Wiener Filter + gradient 0.604 23.52 0.751
11 PnP-DnCNN + gradient 0.600 23.52 0.751
12 ResUNet + gradient 0.583 23.42 0.747
13 PnP-FISTA + gradient 0.519 20.34 0.615
Spec Ranges (3 parameters)
Parameter Min Max Unit
pulse_width 72.0 192.0 fs
gdd -350.0 1150.0 fs²
scattering -7.0 23.0 %
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

Two-photon microscopy uses ultrashort pulsed near-infrared laser light (typically 700-1000 nm) to excite fluorophores via simultaneous absorption of two photons, providing intrinsic optical sectioning because excitation only occurs at the focal volume where photon density is sufficiently high. The longer excitation wavelength enables imaging depths of 500-1000 um in scattering tissue (e.g., brain), making it the standard for in vivo neuroscience. The point-spread function is effectively the square of the excitation PSF. Primary degradations include scattering-induced signal loss with depth and wavefront aberrations from tissue inhomogeneity.

Principle

Two-photon excitation uses a pulsed near-infrared laser so that two photons are absorbed simultaneously by a fluorophore, producing fluorescence equivalent to a single photon of half the wavelength. Because absorption depends on the square of intensity, fluorescence is generated only at the tight focus, providing intrinsic optical sectioning without a pinhole. Deep tissue penetration (up to ~1 mm) is achieved due to reduced scattering at NIR wavelengths.

How to Build the System

Install a mode-locked Ti:Sapphire laser (680-1080 nm, ~100 fs pulses, 80 MHz, Coherent Chameleon or Spectra-Physics InSight) on a laser-scanning microscope. Use a high-NA water-dipping objective (25x 1.05 NA or 20x 1.0 NA) for deep imaging. Non-descanned detectors (GaAsP PMTs) collect scattered fluorescence close to the objective for maximum efficiency. Add a Pockels cell for fast power modulation.

Common Reconstruction Algorithms

  • Adaptive background subtraction for in-depth imaging
  • Motion correction and image registration for in-vivo data
  • Suite2p / CaImAn (calcium imaging segmentation and trace extraction)
  • Deep-learning denoising (DeepInterpolation, Noise2Void)
  • Attenuation compensation (exponential depth correction)

Common Mistakes

  • Excessive laser power causing photodamage and heating deep in tissue
  • Pre-chirp not compensated, broadening pulses and reducing two-photon efficiency
  • Crosstalk between emission channels when using multiple fluorophores
  • Brain motion artifacts in in-vivo imaging not corrected
  • Imaging too deep without correcting for signal attenuation with depth

How to Avoid Mistakes

  • Titrate laser power to minimum effective level; monitor for tissue damage signs
  • Use a prism-pair or grating pre-chirp compressor to maintain short pulses at the focus
  • Select well-separated emission spectra and use appropriate dichroics and filters
  • Apply real-time or post-hoc motion correction algorithms (rigid or non-rigid)
  • Use adaptive optics or longer-wavelength excitation (three-photon) for deep tissue

Forward-Model Mismatch Cases

  • The widefield fallback uses a linear Gaussian PSF, but two-photon excitation depends on intensity squared (I^2), producing a much tighter effective PSF — the fallback PSF is 40-60% wider than the true two-photon PSF
  • The widefield model applies uniform illumination, but two-photon intrinsically provides optical sectioning (only the focal volume has sufficient intensity for I^2 absorption) — the out-of-focus background model is fundamentally wrong

How to Correct the Mismatch

  • Use the two-photon operator with the squared PSF: effective_PSF = PSF_excitation^2, which is ~1.4x narrower than the single-photon PSF
  • Model the nonlinear excitation correctly; for deep tissue, include scattering-induced PSF broadening and signal attenuation with depth

Experimental Setup — Signal Chain

Experimental setup diagram for Two-Photon / Multiphoton Microscopy

Experimental Setup

Instrument: Thorlabs Bergamo II / Bruker Ultima Investigator
Objective: XLUMPLFLN 20x / 0.95 NA water immersion (Olympus)
Pixel Size Nm: 330
Excitation Source: Ti:Sapphire laser (Coherent Chameleon, 920 nm)
Pulse Width Fs: 100
Repetition Rate Mhz: 80
Average Power Mw: 30
Dwell Time Us: 2
Imaging Depth Um: 500
Detector: GaAsP PMT (non-descanned)

Key References

  • Denk et al., 'Two-photon laser scanning fluorescence microscopy', Science 248, 73-76 (1990)
  • Helmchen & Denk, 'Deep tissue two-photon microscopy', Nature Methods 2, 932-940 (2005)

Canonical Datasets

  • Allen Brain Observatory two-photon calcium imaging
  • Stringer et al. (2019) mouse V1 two-photon dataset

Spec DAG — Forward Model Pipeline

C(PSF_2P) → D(g, η₃)

C Two-Photon PSF (PSF_2P)
D Non-Descanned PMT (g, η₃)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δτ pulse_width Pulse width broadening (fs) 100 140
ΔGDD gdd Group delay dispersion error (fs²) 0 500
Δμ_s scattering Tissue scattering error (%) 0 10.0

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.