ENHANCING A SPATIAL-FREQUENCY MUTUAL NETWORK BY RESIDUAL LEARNING FOR FACE SUPER-RESOLUTION(RE-SFMNet)

Our experimental code will be published later！

Visual quality comparison

Visual quality comparison on CelebA dataset by the scale of ×8

Visual quality comparison on Helen dataset by the scale of ×8

Quantitative comparisons for ×8 SR on the CelebA and Helen test sets.

Method	CelebA PSNR	CelebA SSIM	Helen PSNR	Helen SSIM
Bicubic	23.58	0.6285	23.88	0.6628
SRCNN	23.93	0.6348	24.72	0.6770
EDSR	26.84	0.7787	26.60	0.7851
FSRNet	26.66	0.7714	26.43	0.7799
DIC	27.37	0.8022	26.94	0.8026
SPARNet	27.42	0.8036	26.95	0.8029
SISN	27.31	0.7978	27.08	0.8083
SFMNet	27.56	0.8047	27.22	0.8141
RE-SFMNet（ours）	27.70	0.8126	27.26	0.8163

Requirement

pytorch 1.12.1 Cuda 11.4

Train

Test

Pretrain Model