Improving image generative models with human interactions

   Abstract