search

zhangchenxu528 / FACIAL

FACIAL: Synthesizing Dynamic Talking Face With Implicit Attribute Learning. ICCV, 2021.
GNU Affero General Public License v3.0
376 stars 83 forks source link

mouth and eyes are still blurry #46

Open czy36mengfei opened 2 years ago

czy36mengfei commented 2 years ago

I ran the model on your side of the way, a total of 50 epochs of training, but the mouth and eyes are still blurry. Especially in the mouth, which is a bit unacceptably blurry. May I ask what part is wrong?

mouth eyes

zhao1f commented 2 years ago

Hi, it seems there is a little blurry when zooming into the detailed regions. I think you could check the training of the final rendering to video procedure, or tuning with more epochs or training data. Also, there is another probable reason that some certain blurriness in the model in this repo is caused by the final usage of well-known pix2pix baseline.. Some cases are also blurred if you zoom in cases.

I think maybe you could tune the final generation network or modify this part with other strong generation models in the final stage, including stylegan and replace them in the overall procedure.