search

TencentARC / GFPGAN

GFPGAN aims at developing Practical Algorithms for Real-world Face Restoration.
Other
35.04k stars 5.8k forks source link

How to train on 1024 x 1024 face images? #534

Open danial880 opened 3 months ago

danial880 commented 3 months ago

Hi, Thanks for your great work.

I have cropped faces of size 1024x1024. Right now, I am slicing the face into 4 equal slices of 512x512, applying face restoration on each slice, and then concatenating these slices.

I want to directly train a model on 1024x1024 images. So I can skip concatenating. What modification can I make to train_gfpgan_v1.yml, to achieve this?

OmlineEditor commented 2 months ago

Have you been able to find a solution to your problem? Share the solution if you have found it.

jkla139 commented 2 months ago

I find his project is work: GFPGAN-1024 #356

luser350 commented 1 month ago

GFPGAN-1024 takes input 512 and gives 1024 as output. I also want input as 1024. @OmlineEditor @danial880 @jkla139 @tuhins @VincentSC any idea?

OmlineEditor commented 2 weeks ago

I also want input as 1024

do you want to have a 1024 photo at the input and the exact same 1024 photo at the output? I don't quite understand what you want to achieve?