Public

EELS — 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
energy_dispersion	-0.002 – 0.004	0.001	eV/channel
zero_loss_shift	-0.3 – 0.6	0.15	eV
aberration	-2.0 – 4.0	1.0	%

Method: CPU_baseline — Mismatch parameter: nominal

22.54 dB

SSIM 0.3460

18.51 dB

SSIM 0.1083

20.11 dB

SSIM 0.1956

Per-scene breakdown (4 scenes)

Scene	PSNR I	SSIM I	PSNR II	SSIM II	PSNR III	SSIM III
scene_00	23.51	0.3680	18.32	0.1039	20.03	0.1989
scene_01	20.04	0.2917	18.86	0.1280	20.03	0.1975
scene_02	23.28	0.3575	18.45	0.0937	20.28	0.1860
scene_03	23.32	0.3669	18.41	0.1076	20.10	0.2001

#	Method	Score	PSNR	SSIM	Consistency	Trust	Source
1	DiffEELS + gradient	0.859	37.82	0.981	0.94	✓ Certified	Gao et al., NeurIPS 2024
2	PhysEELS + gradient	0.842	36.71	0.977	0.92	✓ Certified	Chen et al., Microsc. Microanal. 2024
3	SwinEELS + gradient	0.806	34.4	0.964	0.88	✓ Certified	Wang et al., npj Comput. Mater. 2023
4	TransEELS + gradient	0.783	32.29	0.946	0.91	✓ Certified	Li et al., Ultramicroscopy 2022
5	N2V-EELS + gradient	0.744	29.72	0.912	0.91	✓ Certified	Krull et al., NeurIPS 2019
6	DnCNN-EELS + gradient	0.733	28.93	0.899	0.92	✓ Certified	Kovarik et al., npj Comput. Mater. 2016
7	ICA-EELS + gradient	0.645	24.94	0.8	0.88	✓ Certified	Bosman et al., Ultramicroscopy 2006
8	MLS-EELS + gradient	0.612	23.06	0.733	0.94	✓ Certified	Verbeeck & Van Aert, Ultramicroscopy 2004
9	PowerLaw-EELS + gradient	0.488	18.87	0.543	0.91	✓ Certified	Egerton, EELS in the EM, Springer 2011

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%