Jianzhu Guo
Jianzhu Guo
I think you may need to parse [BFM_UV.mat](https://github.com/cleardusk/3DDFA_V2/blob/master/configs/BFM_UV.mat).
https://github.com/cleardusk/3DDFA_V2/blob/9fdbea1eb97f762221f71f5c76f08f52296c6704/TDDFA.py#L134 But the z is not the real depth, since 3DDFA_V2 adopts the weak-perspective projection (or scaled orthogonal projection).
You can try to refer to: https://github.com/YadiraF/face3d/blob/master/examples/Data/BFM/readme.md
You can refer to [FLAME](https://github.com/TimoBolkart/TF_FLAME) and other head models recently published.
You can use the default Sim3DR, disable the c render. For example, comment out line 13 and uncomment out line 12. https://github.com/cleardusk/3DDFA_V2/blob/fa8dfc479b46c218e7d375706c673d5823ddb464/demo.py#L12-L13 Accordingly, comment out the build scripts in Line9-11...
Currently, you can refer to the paper for details. The meta-joint part maybe a little hard to implement. I suggest you can use the landmark branch and the short-video-synthesis parts...
Because the mouth region of UV mapping are with hole. I think changing the UV may be fine.
Thanks for your effort!
There is no docker now. Welcome for a PR on it.
The topic you are interested may be ”face frontalization“, you can search the keyword using Google Scholar.