CodeFormer icon indicating copy to clipboard operation
CodeFormer copied to clipboard

How to config for training with resolution 1024

Open hongdthaui opened this issue 1 year ago • 4 comments
trafficstars

Hi, I want to training the model with input datasets resolution 1024, so where do I need to configure it? Thank you!

hongdthaui avatar Dec 02 '23 09:12 hongdthaui

https://github.com/sczhou/CodeFormer/blob/8392d0334956108ab53d9439c4b9fc9c4af0d66d/options/VQGAN_512_ds32_nearest_stage1.yml#L17

and

https://github.com/sczhou/CodeFormer/blob/8392d0334956108ab53d9439c4b9fc9c4af0d66d/options/VQGAN_512_ds32_nearest_stage1.yml#L18

Roller44 avatar Dec 07 '23 01:12 Roller44

@Roller44 Ro Thank for reply, i think i can use file latent_gt_code1024.pth and vqgan_code1024.pth without having retrain stage 1. I want change config in stage 2 and 3 to be training 1024. i edited https://github.com/sczhou/CodeFormer/blob/8392d0334956108ab53d9439c4b9fc9c4af0d66d/options/CodeFormer_stage2.yml#L17 and https://github.com/sczhou/CodeFormer/blob/8392d0334956108ab53d9439c4b9fc9c4af0d66d/options/CodeFormer_stage2.yml#L18 but not ok, it will give an error when initializing

hongdthaui avatar Dec 18 '23 02:12 hongdthaui

@hongdthaui Maybe it is better to show your error messages.

Roller44 avatar Dec 26 '23 06:12 Roller44

Thank for reply, i re-training with 512 resolution but result not good

hongdthaui avatar Jan 09 '24 16:01 hongdthaui