Public

Confocal 3D — Public Tier

(5 scenes)

Full-access development tier with all data visible.

Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

Reconstructed signals (x_hat) and corrected spec as HDF5.

✓

True spec visible — use these exact values for Scenario III oracle reconstruction.

Parameter	Spec Range	True Value	Unit
z_step	-50.0 – 100.0	25.0	nm
spherical_aberr	-0.1 – 0.2	0.05	waves
refractive_index	1.505 – 1.535	1.52

Method: CPU_baseline — Mismatch parameter: nominal

20.05 dB

SSIM 0.2448

17.59 dB

SSIM 0.1380

7.05 dB

SSIM 0.0049

Per-scene breakdown (4 scenes)

Scene	PSNR I	SSIM I	PSNR II	SSIM II	PSNR III	SSIM III
scene_00	17.89	0.0189	17.75	0.0093	8.40	0.0032
scene_01	22.78	0.3272	20.41	0.1455	8.58	0.0017
scene_02	22.25	0.5892	16.29	0.3759	5.62	0.0117
scene_03	17.27	0.0438	15.91	0.0213	5.60	0.0032

#	Method	Score	PSNR	SSIM	Consistency	Trust	Source
1	Restormer-3D + gradient	0.851	37.5	0.98	0.92	✓ Certified	Zamir et al., CVPR 2022 (3D adapted)
2	DiffusionMicro + gradient	0.844	37.28	0.979	0.9	✓ Certified	Gao et al., Nat. Methods 2024
3	U-Net-3D + gradient	0.820	34.89	0.967	0.92	✓ Certified	Çiçek et al., MICCAI 2016
4	SwinIR-3D + gradient	0.817	35.47	0.971	0.87	✓ Certified	Liang et al., ICCV 2021 (3D adapted)
5	CARE + gradient	0.781	32.73	0.95	0.87	✓ Certified	Weigert et al., Nat. Methods 2018
6	Noise2Void + gradient	0.764	31.52	0.937	0.87	✓ Certified	Krull et al., CVPR 2019
7	IRCNN-Confocal + gradient	0.740	29.69	0.912	0.89	✓ Certified	Zhang et al., CVPR 2017
8	Wiener-3D + gradient	0.669	25.83	0.827	0.9	✓ Certified	Wiener, 1942
9	Richardson-Lucy + gradient	0.662	25.04	0.803	0.95	✓ Certified	Richardson, J. Opt. Soc. Am. 1972

y H_ideal spec_ranges x_true true_spec

Format: HDF5

Scenes: 5

0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

PSNR: 40% SSIM: 40% Consistency: 20%