Giters

chiehwangs / gaussian-head

Official repository for 'GaussianHead: High-fidelity Head Avatars with Learnable Gaussian Derivation'

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Custom data processing.

BPAWD opened this issue · comments

commented

Hi, thanks for open-sourcing this good work. How to apply this method to a custom dataset? Could you provide the data processing code?

commented

Thanks for your prompt reply! Any tips for adjusting the parameters for other 3DMM like FLAME? BTW, do you have a timeline for releasing the training code? Thanks.

commented

Thanks for your prompt reply! Any tips for adjusting the parameters for other 3DMM like FLAME? BTW, do you have a timeline for releasing the training code? Thanks.

Hi,

  1. I think using expressions from other 3DMMs is kind of an experimental thing, not really getting into theoretical changes. So, you can try tweaking the network complexity or using specific training tricks. Unfortunately, I haven't experimented a lot in this area, just tested out some feasibility. In past research, many projects have used diverse expression sources, like 'NerFace' using face2face, 'IMavatar' using FLAME, and so on. You can check those out to see how they play with expressions differently.

  2. The release date for our train.py is still up in the air, but I don't think it will be too long.

commented

Hi, thanks for open-sourcing this good work. How to apply this method to a custom dataset? Could you provide the data processing code?

Have you made your own dataset?