Table 2.

Omitting the fine-tuning stage of our optimization process at test time (see Sect. 7.1) leads to significantly lower scores (“Fit Only”)

	View synthesis			View synthesis + Relighting
	Fit Only	Single z	Full Model	w/o ${\mathcal{L}}_{\text{O}}$	w/o $\mathcal{R}$	Full model
PSNR	22.39	24.53	26.67	20.71	20.81	22.80
SSIM	0.71	0.82	0.82	0.61	0.72	0.76

We evaluate the two latent space design choices (“Single z”). We observe that using a disentangled latent space design (see Sect. 7.4) leads to improved performance, mainly attributed to a better face prior representation that helps in generalization. Our evaluations show that using ${\mathcal{L}}_{\text{O}}$ instead of MSE loss (“w/o ${\mathcal{L}}_{\text{O}}$”) to supervise HDR improves the performance of our method (see bold). We quantitatively demonstrate the significance Reflectance Network (“w/o $\mathcal{R}$”). Clearly having a dedicated Reflectance Network improves the relighting quality