astra-vision / CoMoGAN

CoMoGAN: continuous model-guided image-to-image translation. CVPR 2021 oral.
Apache License 2.0
180 stars 14 forks source link

question about codes #13

Closed DZY-cybe closed 2 years ago

DZY-cybe commented 2 years ago

Thanks for your codes!I have one question about restart a training.In the README.md, I seem to be able to use: python train.py --id train_ID --path_data path/to/waymo/training/dir --gpus 0.But when I use the pretrained model, it builds a new version.Is it right?

fabvio commented 2 years ago

It shouldn't. Are you trying to finetune a model?

DZY-cybe commented 2 years ago

Yes.I also want to restart a stopped training.

fabvio commented 2 years ago

I think this is simply due to the pretrained model being in a folder with a different name. If needed, you can bypass Pytorch Lightning checkpointing mechanism and simply load the model with a load_state_dict call.