DarkGeekMS/portrait-synthesis-from-speech

Expand the current set of feature directions through various method

Closed this issue · 0 comments

Expand the set of current feature directions through one or more of the following methods :

  • First method :

    • Generate a set of random faces using StyleGAN2.
    • Classify (or label) the generated faces (manually or using binary classifier).
    • Fit a single feature direction using Logistic Regression or any other classifier or regressor.
  • Second method :

    • Collect some classified faces about a specific feature.
    • Project the collected faces to W.
    • Fit a single feature direction using Logistic Regression or any other classifier or regressor.
  • Third method :

    • Use PCA components proposed in GANSpace.
  • Fourth method :