FTIR Spectroscopic Imaging

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
🥇	PINN-Spectra PINN-Spectra Physics-informed neural network 33.17 dB SSIM 0.954 Checkpoint unavailable	0.780	33.17	0.954	✓ Certified	Physics-informed neural network
🥈	SpectraFormer SpectraFormer Spectroscopy transformer, 2024 32.72 dB SSIM 0.950 Checkpoint unavailable	0.770	32.72	0.950	✓ Certified	Spectroscopy transformer, 2024
🥉	Cascade-UNet Cascade-UNet Physics-informed UNet, 2025 33.0 dB SSIM 0.922 Checkpoint unavailable	0.761	33.0	0.922	✓ Certified	Physics-informed UNet, 2025
4	CDAE CDAE Zhang et al., Sensors 2024 31.5 dB SSIM 0.895 Checkpoint unavailable	0.723	31.5	0.895	✓ Certified	Zhang et al., Sensors 2024
5	U-Net-Spectra U-Net-Spectra Spectral U-Net variant 30.57 dB SSIM 0.925 Checkpoint unavailable	0.722	30.57	0.925	✓ Certified	Spectral U-Net variant
6	DiffusionSpectra DiffusionSpectra Zhang et al., 2024 30.09 dB SSIM 0.918 Checkpoint unavailable	0.711	30.09	0.918	✓ Certified	Zhang et al., 2024
7	ScoreSpectra ScoreSpectra Wei et al., 2025 29.49 dB SSIM 0.909 Checkpoint unavailable	0.696	29.49	0.909	✓ Certified	Wei et al., 2025
8	PnP-DnCNN PnP-DnCNN Zhang et al., 2017 27.9 dB SSIM 0.800 Try in SpecLab →	0.615	27.9	0.800	✓ Certified	Zhang et al., 2017
9	Baseline Correction Baseline Correction Polynomial fitting baseline 25.03 dB SSIM 0.803 Try in SpecLab →	0.569	25.03	0.803	✓ Certified	Polynomial fitting baseline
10	SVD SVD Singular Value Decomposition 24.99 dB SSIM 0.802 Try in SpecLab →	0.568	24.99	0.802	✓ Certified	Singular Value Decomposition
11	SG-ALS SG-ALS Savitzky-Golay + ALS baseline 24.3 dB SSIM 0.670 Try in SpecLab →	0.490	24.3	0.670	✓ Certified	Savitzky-Golay + ALS 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
🥇	PINN-Spectra + gradient PINN-Spectra + gradient Physics-informed neural network Score 0.708 Correct & Reconstruct →	0.708	0.781 31.81 dB / 0.941	0.693 27.82 dB / 0.877	0.650 25.95 dB / 0.831	✓ Certified	Physics-informed neural network
🥈	Cascade-UNet + gradient Cascade-UNet + gradient Physics-informed UNet, 2025 Score 0.657 Correct & Reconstruct →	0.657	0.776 31.41 dB / 0.936	0.615 23.91 dB / 0.765	0.579 22.41 dB / 0.707	✓ Certified	Physics-informed UNet, 2025
🥉	SpectraFormer + gradient SpectraFormer + gradient Spectroscopy transformer, 2024 Score 0.656 Correct & Reconstruct →	0.656	0.751 30.49 dB / 0.924	0.660 26.34 dB / 0.841	0.558 22.37 dB / 0.706	✓ Certified	Spectroscopy transformer, 2024
4	CDAE + gradient CDAE + gradient Zhang et al., Sensors 2024 Score 0.639 Correct & Reconstruct →	0.639	0.755 30.3 dB / 0.921	0.614 23.51 dB / 0.751	0.549 21.47 dB / 0.667	✓ Certified	Zhang et al., Sensors 2024
5	U-Net-Spectra + gradient U-Net-Spectra + gradient Spectral U-Net variant Score 0.629 Correct & Reconstruct →	0.629	0.718 28.76 dB / 0.896	0.625 24.49 dB / 0.785	0.544 21.34 dB / 0.661	✓ Certified	Spectral U-Net variant
6	DiffusionSpectra + gradient DiffusionSpectra + gradient Zhang et al., 2024 Score 0.627 Correct & Reconstruct →	0.627	0.703 27.5 dB / 0.870	0.623 24.39 dB / 0.782	0.554 22.15 dB / 0.696	✓ Certified	Zhang et al., 2024
7	PnP-DnCNN + gradient PnP-DnCNN + gradient Zhang et al., 2017 Score 0.594 Correct & Reconstruct →	0.594	0.666 26.04 dB / 0.833	0.597 23.26 dB / 0.741	0.520 20.43 dB / 0.619	✓ Certified	Zhang et al., 2017
8	Baseline Correction + gradient Baseline Correction + gradient Polynomial fitting baseline Score 0.573 Correct & Reconstruct →	0.573	0.591 22.64 dB / 0.717	0.560 22.24 dB / 0.700	0.567 22.4 dB / 0.707	✓ Certified	Polynomial fitting baseline
9	SVD + gradient SVD + gradient Singular Value Decomposition Score 0.558 Correct & Reconstruct →	0.558	0.596 22.97 dB / 0.730	0.552 21.4 dB / 0.664	0.526 21.27 dB / 0.658	✓ Certified	Singular Value Decomposition
10	SG-ALS + gradient SG-ALS + gradient Savitzky-Golay + ALS baseline Score 0.553 Correct & Reconstruct →	0.553	0.613 23.26 dB / 0.741	0.550 21.71 dB / 0.677	0.495 20.12 dB / 0.604	✓ Certified	Savitzky-Golay + ALS baseline
11	ScoreSpectra + gradient ScoreSpectra + gradient Wei et al., 2025 Score 0.546 Correct & Reconstruct →	0.546	0.699 27.58 dB / 0.872	0.510 19.96 dB / 0.597	0.430 17.32 dB / 0.466	✓ Certified	Wei et al., 2025

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	PINN-Spectra + gradient	0.781	31.81	0.941
2	Cascade-UNet + gradient	0.776	31.41	0.936
3	CDAE + gradient	0.755	30.3	0.921
4	SpectraFormer + gradient	0.751	30.49	0.924
5	U-Net-Spectra + gradient	0.718	28.76	0.896
6	DiffusionSpectra + gradient	0.703	27.5	0.87
7	ScoreSpectra + gradient	0.699	27.58	0.872
8	PnP-DnCNN + gradient	0.666	26.04	0.833
9	SG-ALS + gradient	0.613	23.26	0.741
10	SVD + gradient	0.596	22.97	0.73
11	Baseline Correction + gradient	0.591	22.64	0.717

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
wavenumber_calibration	-0.4	0.8	cm^-1
water_vapor_absorption	-0.15	0.15	-
detector_nonlinearity	-1.0	2.0	-
atr_crystal_ri_error	-0.2	0.4	-

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	PINN-Spectra + gradient	0.693	27.82	0.877
2	SpectraFormer + gradient	0.660	26.34	0.841
3	U-Net-Spectra + gradient	0.625	24.49	0.785
4	DiffusionSpectra + gradient	0.623	24.39	0.782
5	Cascade-UNet + gradient	0.615	23.91	0.765
6	CDAE + gradient	0.614	23.51	0.751
7	PnP-DnCNN + gradient	0.597	23.26	0.741
8	Baseline Correction + gradient	0.560	22.24	0.7
9	SVD + gradient	0.552	21.4	0.664
10	SG-ALS + gradient	0.550	21.71	0.677
11	ScoreSpectra + gradient	0.510	19.96	0.597

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
wavenumber_calibration	-0.48	0.72	cm^-1
water_vapor_absorption	-0.15	0.15	-
detector_nonlinearity	-1.2	1.8	-
atr_crystal_ri_error	-0.24	0.36	-

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	PINN-Spectra + gradient	0.650	25.95	0.831
2	Cascade-UNet + gradient	0.579	22.41	0.707
3	Baseline Correction + gradient	0.567	22.4	0.707
4	SpectraFormer + gradient	0.558	22.37	0.706
5	DiffusionSpectra + gradient	0.554	22.15	0.696
6	CDAE + gradient	0.549	21.47	0.667
7	U-Net-Spectra + gradient	0.544	21.34	0.661
8	SVD + gradient	0.526	21.27	0.658
9	PnP-DnCNN + gradient	0.520	20.43	0.619
10	SG-ALS + gradient	0.495	20.12	0.604
11	ScoreSpectra + gradient	0.430	17.32	0.466

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
wavenumber_calibration	-0.28	0.92	cm^-1
water_vapor_absorption	-0.15	0.15	-
detector_nonlinearity	-0.7	2.3	-
atr_crystal_ri_error	-0.14	0.46	-

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 → Σ → D

M Modulation

Σ Summation

D Detector

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
w_c	wavenumber_calibration	Wavenumber calibration (cm^-1)	0.4
w_v	water_vapor_absorption	Water vapor absorption (-)	0.0
d_n	detector_nonlinearity	Detector nonlinearity (-)	1.0
a_c	atr_crystal_ri_error	ATR crystal RI error (-)	0.2

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.