Resume model training instead of start from scratch

[Cioscos]

Hi! I would like to know if it's possible for you to implement a training resuming because it's impossible to keep a training for very long time, and we would like to continue training of other model to improve them. Another thing is that --pretrain arg in the argument parser is not used. Will pre-trained models ever be available for gpen? Will the option to use pre-trained models ever exist? Another courtesy is to specify a little better, perhaps with the "help" arg, what each argument of the argument parser does. Thank you in advance.

[seranus]

resume training would be very useful

[su161021]

请问大佬 训练好的模型怎么进行infer测试？

[hukangli]

借个楼，请问大佬，训练集选择有什么要求吗？能用GFPGAN生成的高清图片作为训练集吗？

[yangxy]

Hi! I would like to know if it's possible for you to implement a training resuming because it's impossible to keep a training for very long time, and we would like to continue training of other model to improve them. Another thing is that --pretrain arg in the argument parser is not used. Will pre-trained models ever be available for gpen? Will the option to use pre-trained models ever exist? Another courtesy is to specify a little better, perhaps with the "help" topic, what each argument of the argument parser does. Thank you in advance.

Updated.

[zhangyunming]

just fintune, where is the resume?

[jinwonkim93]

just fintune, where is the resume?

I was able to resume by changing few things.

https://github.com/yangxy/GPEN/pull/160