StyleGAN2 : Analyzing and Improving the Image Quality of StyleGAN

• 제목 : Analyzing and Improving the Image Quality of StyleGAN
• 아카이브 ID : 1912.04958
• 깃허브 코드 :
• 저자 : T.Karras, S.Laine, J.Lehtinen 등
• 발표 년도 : 2019

Overview

• We expose and analyze several of STYLEGAN’s characteristic artifacts, and propose changes in both model architecture and training methods.
  • Redesign generator normalization
  • Revisit progressive growing
  • Regularize the generator for good conditioning
    • Improve image quality
    • Easier generator invertibility

Normalization Artifacts

• exhibit characteristic blob-shaped artifacts that resemble water droplets
  • starts to appear around 64×64 resolution, is present in all feature maps, and becomes progressively stronger
• Problem : AdaIN operation
  • normalizes the mean and variance of each feature map separately, thereby potentially destroying any information found in the magnitudes of the features relative to each other.

    Written with StackEdit.