Labels4Free: Unsupervised Segmentation using StyleGAN

  • 2021-03-27 18:59:22
  • Rameen Abdal, Peihao Zhu, Niloy Mitra, Peter Wonka
  • 52


We propose an unsupervised segmentation framework for StyleGAN generatedobjects. We build on two main observations. First, the features generated byStyleGAN hold valuable information that can be utilized towards trainingsegmentation networks. Second, the foreground and background can often betreated to be largely independent and be composited in different ways. For oursolution, we propose to augment the StyleGAN2 generator architecture with asegmentation branch and to split the generator into a foreground and backgroundnetwork. This enables us to generate soft segmentation masks for the foregroundobject in an unsupervised fashion. On multiple object classes, we reportcomparable results against state-of-the-art supervised segmentation networks,while against the best unsupervised segmentation approach we demonstrate aclear improvement, both in qualitative and quantitative metrics.


Quick Read (beta)

loading the full paper ...