search

ashawkey / RAD-NeRF

Real-time Neural Radiance Talking Portrait Synthesis via Audio-spatial Decomposition
MIT License
878 stars 153 forks source link

Function of Individual codes #40

Closed sora158 closed 1 year ago

sora158 commented 1 year ago

I find network has a input c, and it refers to individual codes. However , I can't find individual code in your paper. Can you explain what's the meaning of this input? Thank you!

sora158 commented 1 year ago

In nerf/render.py, Line 199 

   if self.individual_dim > 0:
        if self.training:
            ind_code = self.individual_codes[index]
        # use a fixed ind code for the unknown test data.
        else:
            ind_code = self.individual_codes[0]
    else:
        ind_code = None
yerfor commented 1 year ago

I suppose it might be a common practice in the field of dynamic NeRF to improve image fidelity (PSNR, etc). But it seems no significant improvement in this case though.

sora158 commented 1 year ago

I suppose it might be a common practice in the field of dynamic NeRF to improve image fidelity (PSNR, etc). But it seems no significant improvement in this case though.

Thanks for replying! I'm new to Dynamic nerf ^^. Also, I follow up your research of Geneface, your demo which uses DiffSinger looks wonderful!

yerfor commented 1 year ago

Thanks~