A great project, the source code author worked hard.

Fictionarry / ER-NeRF

[ICCV'23] Efficient Region-Aware Neural Radiance Fields for High-Fidelity Talking Portrait Synthesis

https://fictionarry.github.io/ER-NeRF/

MIT License

894 stars 124 forks source link

Open langzizhixin opened 8 months ago

langzizhixin commented 8 months ago

Changing "face_parsing" and "face_tracking" to the algorithms "Segment Anything" and "Track Anything", that were just released this year will git greatly improve the effect. https://github.com/Fictionarry/ER-NeRF/assets/74521932/3549d467-22fd-41b2-9ce2-a247c920cd48

langzizhixin commented 8 months ago

Please clarify. My training video is about digital people. Because I am relatively lazy, I didn't go to find a model to shoot on my own.

deligenters commented 8 months ago

May I ask how the "Track Anything" model fits into 3DMM modeling and generates the transformation matrix

husthzy commented 8 months ago

the "Track Anything" is different from 3DMM face tracker, so I think it can not replace the face tracker