Public

4D-STEM — Public Tier

(3 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
camera_length	-2.0 – 4.0	1.0	%
center_offset	-1.0 – 2.0	0.5	pixels
elliptical_distortion	-0.005 – 0.01	0.0025

Method: CPU_baseline — Mismatch parameter: nominal

19.81 dB

SSIM 0.4840

17.13 dB

SSIM 0.0865

19.14 dB

SSIM 0.1862

Per-scene breakdown (4 scenes)

Scene	PSNR I	SSIM I	PSNR II	SSIM II	PSNR III	SSIM III
scene_00	19.40	0.4889	16.86	0.0892	19.02	0.1932
scene_01	20.39	0.4977	17.19	0.0846	19.00	0.1825
scene_02	20.11	0.4612	17.55	0.0813	19.72	0.1792
scene_03	19.34	0.4883	16.90	0.0911	18.84	0.1898

#	Method	Score	PSNR	SSIM	Consistency	Trust	Source
1	PhysED + gradient	0.839	36.32	0.975	0.93	✓ Certified	Chen et al., Nat. Commun. 2024
2	DiffED + gradient	0.837	36.95	0.978	0.88	✓ Certified	Gao et al., NeurIPS 2024
3	SwinED + gradient	0.801	33.87	0.96	0.89	✓ Certified	Wang et al., npj Comput. Mater. 2023
4	TransED + gradient	0.780	32.33	0.946	0.89	✓ Certified	Li et al., Nat. Commun. 2022
5	PhaseGAN-ED + gradient	0.767	30.77	0.928	0.94	✓ Certified	Zimmermann et al., Sci. Adv. 2021
6	DnCNN-ED + gradient	0.721	28.14	0.884	0.93	✓ Certified	Cherukara et al., npj Comput. Mater. 2018
7	MicroED + gradient	0.641	24.91	0.799	0.86	✓ Certified	Shi et al., eLife 2013
8	PEDT + gradient	0.566	21.76	0.68	0.88	✓ Certified	Kolb et al., Ultramicroscopy 2007
9	Direct-Methods + gradient	0.518	19.67	0.583	0.94	✓ Certified	Hauptman & Karle, Nobel Prize 1985

y H_ideal spec_ranges x_true true_spec

Format: HDF5

Scenes: 3

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

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