Closed Ziad-Mohamedd closed 3 years ago
The inference stage does not need lip-sync_expert.pth
how come ??
The inference.py
script is intended for generating talking head frames. As you can see in figure 2 of the paper it is the Generator that generates talking head frames. There are two pre-trained versions of this Generator wav2lip.pth
and wav2lip_gan.pth
. In contrast, the Lip Sync Expert generates a sync loss which is used during training to tell the generator if it is doing well. If you are interested in this loss value for a specific video you should not use inference.py
which is intended for generating talking head frames and not a specific loss value.
The
inference.py
script is intended for generating talking head frames. As you can see in figure 2 of the paper it is the Generator that generates talking head frames. There are two pre-trained versions of this Generatorwav2lip.pth
andwav2lip_gan.pth
. In contrast, the Lip Sync Expert generates a sync loss which is used during training to tell the generator if it is doing well. If you are interested in this loss value for a specific video you should not useinference.py
which is intended for generating talking head frames and not a specific loss value.
thanks so much for your time ,, its much more clear now thank you
python inference.py --checkpoint_path --face --audio
this line works well with wav2lip.pth and wav2lip_gan.pth which are the pretrained model in the checkpoint path but the pretrained model of the discrimnator which is "lipsync_expert.pth" results in error