Table 3.

Quantitative evaluation of the deep-learning-based reconstructions using Fu’s, Arad’s and our MSFA (see Figure 7) on synthetic datasets.

	MSFA	Dataset
		CAVE		Harvard		ICVL
		RMSE	PSNR	RMSE	PSNR	RMSE	PSNR
Noisy (SNR = ∞)	Ours	0.148	24.46	0.091	29.96	0.054	31.12
	Nie’s	0.166	23.76	0.114	27.07	0.068	29.87
	Arad	0.161	23.54	0.119	26.49	0.069	29.89
Noisy (SNR = 30 db)	Ours	0.163	23.08	0.097	27.23	0.059	30.02
	Nie’s	0.172	22.59	0.118	25.72	0.072	27.93
	Arad	0.186	22.47	0.122	25.06	0.072	28.28