Finetuning GFPGAN

[VaidhegiDinesh]

Hi I am using GFPGAN to enhance the quality of surveillance face images in my project of constrained face recognition in surveillance images. I wish to fine tune the model on the paired dataset I created (I created a paired dataset by extracting kernels from surveillance probe images using kernelGAN and applied it to HR gallery images in my dataset. I also added other degradations to the convolved image.) So, now i wish to fine tune the model on this paired dataset.
i noticed there are several versions of GFPGAN available v1.3, v1.4 etc.

1. Which model do you think will be suitable for my usecase of super resolving surveillance face image for improving face recognition? (note: identity preserving is very important as face recognition is involved)
2. And I dont want to train from scratch, over the existing pretrained model, I need to fine tune with my paired dataset, for that where can i get the pretrained weights?