search

EvelynFan / FaceFormer

[CVPR 2022] FaceFormer: Speech-Driven 3D Facial Animation with Transformers
MIT License
778 stars 133 forks source link

get static facial expression after training in my dataset #22

Closed jyangliu closed 2 years ago

jyangliu commented 2 years ago

After training FaceFormer with my dataset,I find that wave2vec2.0 output a constant feature for all audio,so i got a static facial expression. I add Periodic Positional Encoding in the training time. How to fix that?

jyangliu commented 2 years ago

I solve this problem by seting learning_rate=10e-5 because my own dataset is between [0,1] while vocast dataset is between [-0.1,0.1],so i get a to large learning rate.

ujjawalcse commented 1 year ago

@jyangliu @zlinao @EvelynFan , Me too getting the same. Also, couldn't be fixed with the Learning rate tuning. Any suggestions?