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

On the inconsistency between training results and test results #51

Closed banhuML closed 2 years ago

zhangchenxu528 commented 2 years ago

I think you are incorrectly using Obama's information, which shouldn't be used in your testing. 1) Please check the test_A file to make sure the texture of the rendered result is right, not from Obama. 2) Please check whether the pre-trained network name in the face2vid module is the one you trained.