AliaksandrSiarohin / monkey-net

Animating Arbitrary Objects via Deep Motion Transfer
467 stars 81 forks source link

Training on custom data using pretrained weights #36

Closed abhishekcodework closed 2 years ago

abhishekcodework commented 2 years ago

Firstly, great work on maintaining this repo. I was trying to use the vox pretrained weights to train on my custom data using vox-full.yaml. But the training terminates. It shows 0/20 its and then exits. I ran the following command (custom.yaml is based on vox-full.yaml): CUDA_VISIBLE_DEVICES=0 python run.py --config config/custom.yaml --checkpoint log/custom/vox-cpk.pth.tar Also, thank you

AliaksandrSiarohin commented 2 years ago

Probably, you need to use 2 gpus or reduce the batch size.

abhishekcodework commented 2 years ago

I am able to train without providing --checkpoint with the same config but I get the follwoing error while trying to use vox pretrained weights: ss

AliaksandrSiarohin commented 2 years ago

There is no errors, by default it continues from the last epoch. If it already trained than it will be trained only for one epoch. Increase the number of epochs in config and adjust the lr schedule.

abhishekcodework commented 2 years ago

thank you for the help. Closing this issue now