search

ShenZheng2000 / TPSeNCE

TPSeNCE for image rain generation, deraining, and object detection.
82 stars 0 forks source link

finetune model #13

Closed reza-nouri-9696 closed 1 week ago

reza-nouri-9696 commented 1 week ago

can you guide me about finetune the model. I prepared data and downloaded boreas weights, what should I do in command line? that was a little complex and I cant relate with arguments

ShenZheng2000 commented 1 week ago

Please check this github issue: https://github.com/taesungp/contrastive-unpaired-translation/issues/29

reza-nouri-9696 commented 1 week ago

really thank you, that was so useful but, if my --epoch was latest what should I set --epoch_count? you know, in fact I wanna use boreas_snowy pretrained weight that was access to us by you on your github. Im trying finetune that weight

ShenZheng2000 commented 1 week ago

You can probably try this

--continue_train --epoch_count 1 --epoch latest
reza-nouri-9696 commented 1 week ago

Thank you very much for the quick response I ran this command: !python train.py --dataroot "/content/data/boreas_snowy" --name "boreas_snowy" --checkpoints_dir "/content/weight" --continue_train --epoch_count 1 --epoch latest but the model didnt know loss_T: AttributeError: 'CUTModel' object has no attribute 'loss_T'. Did you mean: 'loss_D'?

ShenZheng2000 commented 1 week ago

It looks like the issue is that lambda_TRI was not assigned a value, causing loss_T to be undefined.

Please refer to this prototype for training to avoid potential errors:(https://github.com/ShenZheng2000/TPSeNCE/blob/master/train.sh).