Where are the path_model and path_opt files when training from scratch? about lcfcn HOT 8 CLOSED

did you try running the checkpoints/download.sh script? It should download those checkpoints.

from lcfcn.

I did. The downloaded folder only had the best_model.pth and history.json files.

from lcfcn.

This is what was provided to download Trancos checkpoints.
https://www.dropbox.com/sh/rms4dg5autwtpnf/AADQBOr1ruFsptbqG_uPt_zCa?dl=0

And it only have the best_model.pth and history.json files.

What am I missing? Can you please help?

from lcfcn.

Hi sorry about that. Can you use best_model.pth to load the model? it should be equivalent to trancos_ResFCN.pth. You can rename the best_model.pth to trancos_ResFCN.pth.

from lcfcn.

Why do you need the checkpoints if you are training from scratch?

from lcfcn.

Sorry I'm very new to deep learning. I thought since main.py needs all arguments path_history, path_model, path_opt, path_best_model for training that those files were necessary. Do I not need them?

from lcfcn.

those paths are there in order to specify where they should be saved, but they don't have to exist. You don't need them.
The image below is part of train.py. It shows that if the path does not exist, then it won't be loaded.

from lcfcn.

Thank you!

from lcfcn.