search

zhanglonghao1992 / One-Shot_Free-View_Neural_Talking_Head_Synthesis

Pytorch implementation of paper "One-Shot Free-View Neural Talking-Head Synthesis for Video Conferencing"
Other
735 stars 141 forks source link

pickle file #33

Open laleh-samadfam opened 2 years ago

laleh-samadfam commented 2 years ago

Hello,

Is the .pkl file for the last checkpoint going to be uploaded? Also, can you provide the information about the number of GPUs and the time you spent for training the model?

Thank you

zhanglonghao1992 commented 2 years ago

@laleh-samadfam Hi~ What you mean 'the .pkl file for the last checkpoin'. Is that Vox-512? I'm not going to train the model at 512 resolution for the time being cause it consumes too much memory and time. Face Restoration algorithms (like GPEN) can be used for post-processing to improve the resolution.

laleh-samadfam commented 2 years ago

Thank you for your answer,

I mean for vox-256. I believe the uploaded "Vox-256-New" does not include discriminator or hopenet weights. Am I misunderstanding something?

zhanglonghao1992 commented 2 years ago

@laleh-samadfam I didn't release the hopenet cause it's not used for inference. I just borrow the model and weight from this repo. The discriminator weight is saved in the .pkl file, referring to line 87-94 in train.py.