Beyond Face Rotation Global and Local Perception GAN for Photorealistic and Identity Preserving Frontal View Synthesis

deal with pose variations

• hand-crafted features: find local distortion, metric learning
• deep-learning methods: tradeoff btw. invarience and discriminability
  • blurry and loss of details
• ill-defined prob.(good point here from optimization’s point of view)
  • failure to provide with prior and constraints lead to blurred images
  • we use GAN as constrains
    • GAN outputs a grid of 0/1: focus on specific region

two-path

• local path for key part

loss

• pixelwise loss
• symmetry loss
• perceptual loss as identity preserving constrains