Question about Test FaceSR

[EthanLiang99]

Great Work! Could you tell me what is the difference between --init-img and --ref-img in Test FaceSR? In sr_val_ddpm_text_T_vqganfin_facerefersampling.py, the default values of --init-img and --ref-img are both /mnt/lustre/share/jywang/dataset/ImageSR/RealSRSet/. It seems like they are the same dataset? Or, --ref-img includes cropped and aligned face images? thx~

[IceClear]

Hi. Thanks for your interest.
The default settings should be a typo.
The --init-img refers to the LR input and the --ref-img refers to a preprocess image, e.g., SR images from an existing SR method such as CodeFormer.

[EthanLiang99]

Thanks for your quick response:)
Will it be better to take the initial restoration with BFR methods (e.g., CodeFormer) before feeding images into FaceSR?

[IceClear]

We observe improvements in natural images under severe degradations in such way, but not sure if it works for face SR.

[EthanLiang99]

OK, I will try it. Thx~

[hangxiaotian]

CodeFormer

Do I need to write the address for the scripts/generate_vqgan_data.py.