search

allenai / satlas-super-resolution

Apache License 2.0
190 stars 24 forks source link

Continuing the training where it stopped #20

Closed yunseok624 closed 4 months ago

yunseok624 commented 4 months ago

What do I have to change in the .yml file to continue where it left off? Just give the path to the saved checkpoint file? Do I also have to change something so that it continued logging in wandb?

piperwolters commented 4 months ago

You will need to add pretrain_network_g and pretrain_network_d arguments to the .yml file to load in weights for both the generator and the discriminator. If you want this to happen automatically, you can specify the --auto_resume flag in the training command.

To continue logging in wandb you'll need to edit the wandb.init() code in the repo. I haven't tested this, but you'd need to set resume=True.

yunseok624 commented 4 months ago

Okay, thank you.