search

NVlabs / imaginaire

NVIDIA's Deep Imagination Team's PyTorch Library
Other
4k stars 448 forks source link

Does fs_vid2vid have a finetune method? #137

Closed cdmc-zrz closed 2 years ago

cdmc-zrz commented 2 years ago

A full fs_vid2vid takes too long to train, I want to finetune this model by training some of the layers to add some side face information, but I don't know how to do it?

mingyuliutw commented 2 years ago

For talking-head, I assume fine-tuning from the voxel-celebrity model might work. For full-body, I assume fine-tuning from the youtube dancing model might work

cdmc-zrz commented 2 years ago

For talking-head, I assume fine-tuning from the voxel-celebrity model might work. For full-body, I assume fine-tuning from the youtube dancing model might work

https://github.com/NVlabs/imaginaire/issues/135 Is there a solution to this problem in fs_vid2vid, there has been a problem loading data according to the process