StyleFaceV icon indicating copy to clipboard operation
StyleFaceV copied to clipboard

Published 20 hours ago • verified publisher icon arthur-qiu

Question about quantitive studies

Open hallwaypzh opened this issue 2 years ago • 1 comments

Thanks for you awesome work; however i am curious:

  1. Number of video samples to compute FVD.
  2. The train-test split of RAVDESS dataset (Since StyleFaceV needs an initial frame).
  3. Will you guys release retrained Di-GAN, StyleGAN-V, MC-HD?

hallwaypzh avatar Oct 13 '22 04:10 hallwaypzh

Hi, 1, 2048. 2, Initial frame can be sampled from the noise directly. 3, No. We retrain them by their default setting and it is easy to reproduce.

arthur-qiu avatar Oct 19 '22 05:10 arthur-qiu