search

Doubiiu / CodeTalker

[CVPR 2023] CodeTalker: Speech-Driven 3D Facial Animation with Discrete Motion Prior
MIT License
537 stars 58 forks source link

the parameters of TCN are fixed, so how to unfreeze #61

Closed Shirley-0708 closed 1 year ago

Shirley-0708 commented 1 year ago

@Doubiiu You mentioned that during training, the parameters of TCN are fixed, whereas the other parts of the model are learnable. Where is the relevant code set for freezing? Can you point it out because I want to unfreeze and make TCN learnable，thanks a lot !

Doubiiu commented 1 year ago

Hi. Thanks for your interest! The corresponding code is L20(self.audio_encoder.feature_extractor._freeze_parameters()) in stage2.py. You may get more information about wav2vec2.0 from the huggingface page / source code or the paper.

Shirley-0708 commented 1 year ago

Thanks for your prompt reply! I have commented out this line self.audio_encoder.feature_extractor._freeze_parameters() and hope it works.