zsyOAOA / ResShift

ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS@2023 Spotlight, TPAMI@2024)
Other
962 stars 50 forks source link

I want to resume from your pre-trained weight. #92

Open H0128Z opened 2 months ago

H0128Z commented 2 months ago

I want to resume from your pre-trained weight, [resshift_realsrx4_s15_v1.pth]

It seems this is ema file, how can I train from pre-trained weight?

below terminal doesnt' work. since it is not dict_keys format

CUDA_VISIBLE_DEVICES=0,1 torchrun --standalone --nproc_per_node=2 --nnodes=1 main.py --cfg_path configs/DNA_5x5micro_2048_10x10crop.yaml --save_dir checkpoint --resume weights/resshift_realsrx4_s15_v1.pth

zsyOAOA commented 2 months ago

Try to pass the ckpt path to the config file. @H0128Z

https://github.com/zsyOAOA/ResShift/blob/journal/configs/realsr_swinunet_realesrgan256_journal.yaml#L29

H0128Z commented 2 months ago

Try to pass the ckpt path to the config file. @H0128Z

https://github.com/zsyOAOA/ResShift/blob/journal/configs/realsr_swinunet_realesrgan256_journal.yaml#L29

ohh,, thanks!