How to get higher resolution?

[jbyu]

Hi @Rudrabha ,

What a great work and thanks for sharing it.
I am wondering

1. How to generate higher resolution output?
2. By cropping face with square area (not resize to 96x96), Will it improve visual quality?

Thank you.

[Rudrabha]

Hello,
Thanks for checking out our work.

1. Right now the model can only generate 96 x 96. This is due to the dataset we use to train Wav2Lip. We are currently trying to train a model on higher resolution data. Once trained, we will release it.
2. Currently, the model needs a 96 x 96 face crop as input. It cannot be changed at the moment.

[zgxiangyang]

Hi @Rudrabha , Is there public higher resolution data available? Could'you please share it.

[prajwalkr]

https://looking-to-listen.github.io/avspeech/download.html

[AlonDan]

https://looking-to-listen.github.io/avspeech/download.html

How do I use these higher quality datasets?
I tried to copy these files to Wav2Lip\checkpoints\ and I renamed them the .csv to .pth
But it doesn't work

Can you please explain how to make these higher resolution models to work?

If there are higher resolution models to download please share the link + how to make it work with wav2lip if it's different than copy them in the checkpoints and use them on the command.

Thanks ahead! :)

[FacundoMartinezCampos]

@AlonDan that is not a model for wav2lip, its a dataset.

[hazemabdelkawy]

May you share the High quality trained model?

[AlonDan]

May you share the High quality trained model?

Yes please! 💙

[lidachuan211]

May you share the High quality trained model?

[Xuey1nger]

Hi @Rudrabha ,how should I train on a high quality dataset.I change the img_size as 288×288 and restructure the model.But I does not get better result than using 96×96.what shoule I change?