search

MRzzm / DINet

The source code of "DINet: deformation inpainting network for realistic face visually dubbing on high resolution video."
907 stars 168 forks source link

Simplfy the network structure #14

Open huangxin168 opened 1 year ago

huangxin168 commented 1 year ago

Amazing! Thanks for your contribution. To simplfy the network structure, Can we:

  1. use 5 mouth images as refrence images instead of 5 whole face?
  2. Fref concat Fs as input to the AdaAT.
  3. no aligment encoder.
  4. no concat in Inpainting part.
huangxin168 commented 1 year ago

I read the AdaAT module, not sure if 5 history mouth+ curent pose can build a good reference feature map for the AdaAT to deform.

973571898 commented 1 year ago

Hello, have you implemented optimization that you had metioned, how was the effect? @huangxin168