search

yerfor / GeneFacePlusPlus

GeneFace++: Generalized and Stable Real-Time 3D Talking Face Generation; Official Code
MIT License
1.47k stars 215 forks source link

Fuzzy generation result problem #72

Open MtYCNN opened 7 months ago

MtYCNN commented 7 months ago

截屏2024-02-20 16 34 42 Thank you for your outstanding contribution！ On the left is the original video frame, and on the right is the generated result. You can see that it has become much blurry. Is there any solution？

Net-Maker commented 6 months ago

Same problem,the hair cannot be represented well.

yunyu commented 6 months ago

I suspect the root cause is because the 3dmm landmark fit is slightly off. There appears to be some temporal smoothing (possibly caused by the euler/trans parameters being trained in batch) going on here, and the radnerf is trained on slightly erroneous alignments which may lead to poor convergence (seen this in previous nerf work).

Ezrealz commented 6 months ago

image image The same problem reasoning results in a fuzzier result than the original video