Table 1 Average Gaussian fitting results of different training and evaluation protocols using multiple frames from the test set containing CDW signals

	μ_h (×10²)	μ_k (×10²)	μ_ℓ (×10)	σ_h (×10²)	σ_k (×10²)	σ_ℓ (×10)	SRBR_h	SRBR_k	SRBR_ℓ
LC	0.66 (05)	1.94 (13)	2.48 (37)	0.18 (04)	1.17 (13)	2.85 (54)	0.39 (07)	0.41 (04)	0.31 (07)
IRUNet
Poisson → Poisson Poisson → Exp. Exp. → Exp.	0.29 (03) 0.75 (04) 0.19 (03)	0.56 (09) 0.87 (11) 0.65 (07)	1.08 (12) 4.17 (14) 1.47 (12)	0.14 (02) 0.31 (04) 0.31 (03)	0.75 (10) 1.65 (13) 0.69 (08)	1.03 (15) 1.37 (17) 1.50 (15)	0.90 (10) 1.02 (25) 1.41 (24)	0.96 (03) 0.95 (13) 1.00 (02)	1.13 (04) 0.97 (04) 1.21 (05)
VDSR
Poisson → Poisson Poisson → Exp. Exp. → Exp.	0.38 (02) 0.46 (03) 0.32 (02)	0.58 (08) 0.72 (14) 0.63 (08)	1.01 (11) 2.62 (11) 1.11 (11)	0.14 (02) 0.19 (02) 0.16 (02)	0.78 (09) 1.28 (18) 0.73 (08)	1.23 (14) 1.18 (14) 0.95 (14)	0.95 (08) 0.94 (07) 0.97 (09)	1.05 (02) 1.15 (03) 1.09 (02)	1.22 (05) 1.24 (05) 1.25 (04)

The first column indicates the used training and evaluation methodology, for example, training on artificial Poisson noise and evaluation on experimental noise (Poisson → Exp.). Values given for training and evaluation on experimental noise are additionally highlighted in bold for visual guidance. The Gaussian peak position μ_α and s.d. σ_α with reciprocal space direction α = (h, k, ℓ) are given as the mean absolute error between the Gaussian parameter obtained from the denoised and the one obtained from the HC signal (the lower the better). Values for the SRBR_α are given as the absolute ratio of the Gaussian parameter obtained from the denoised and the one obtained from the HC signal (the higher the better). Values for μ_α as well as values for σ_α are scaled as indicated. Because of the broader peak in the ℓ direction, a scaling of 10 for μ_ℓ and σ_ℓ has been chosen over a scaling of 100.

Quick links

Search