Mitigating hallucination with non-adversarial strategies for image-to-image translation in solar physics

Open Access

Table A.1

Quantitative comparison of I2IwFiLM and DiffI2I.

		DiffI2I	I2IxFiLM (Ours)
WL-to-CaIIK	RMSE ↓	14.53 ± 7.20	14.44 ± 7.07
	PS/N [dB] ↑	25.88 ± 4.18	25.90 ± 4.12
	SSIM ↑	0.80 ± 0.09	0.80 ± 0.09
	MS-SSIM ↑	0.80 ± 0.11	0.80 ± 0.11
EUV-to-Mag	RMSE ↓	13.09 ± 4.67 × 10⁻³	13.23 ± 5.06 × 10⁻³
	PS/N [dB] ↑	38.19 ± 3.03	38.10 ± 2.98
	SSIM ↑	0.97 ± 0.02	0.96 ± 0.03
	MS-SSIM ↑	0.97 ± 0.02	0.96 ± 0.03

Notes. Comparison is performed for both whitelight-to-calcium translation and EUV-to-Magnetogram translation tasks. Both models implement FiLM feature modulation, with DiffI2I generating guidance vectors $\hat{E^{P}}$ $\[\widehat{\mathcal{E}^{\mathcal{P}}}\]$ using a diffusion process with random seeds, while I2IwFiLM uses direct mapping from input image.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.