search

eeskimez / emotalkingface

The code for the paper "Speech Driven Talking Face Generation from a Single Image and an Emotion Condition"
MIT License
161 stars 29 forks source link

Pretrain emotion disc #7

Closed sahilg06 closed 2 years ago

sahilg06 commented 2 years ago

The classification-loss while pre-training the emotion-disc should go down to what value?

eeskimez commented 2 years ago

Hi,

I don't remember the exact value, but the main idea is that the discriminator should learn to distinguish the emotions before the actual training. You can train it for a small number of epochs (5-7) as an initilization, which should be enough. Just make sure that the loss value is decreasing over time, otherwise, there might be a problem with the data.

Hope this helps!

Best, Emre

sahilg06 commented 2 years ago

Thanks, @eeskimez!

sahilg06 commented 2 years ago

@eeskimez, are the weights of emotion-disc updated during actual training? or they are fixed after pre-training? As if we will update its parameters during actual training, it might focus on visual artifacts instead of emotion.