21.0 Inpainting🔗⧉

Contents of this part

• 21.1 Inpainting, texture synthesis, and object removal
  • 21.1.1 Inpainting as filling unmeasured pixels: the spectrum of priors
  • 21.1.2 PDE / diffusion-based inpainting
  • 21.1.3 Texture synthesis (Efros–Leung; Efros–Freeman quilting)
  • 21.1.4 Exemplar inpainting: clone, healing brush, and object removal (Criminisi)
  • 21.1.5 Data-driven scene completion (Hays & Efros)
  • 21.1.6 Deep inpainting (context encoders → partial/gated conv → diffusion)
  • 21.1.7 Highlight / specular recovery
  • 21.1.8 Epitomes: a compact patch model
• 21.2 Patch match
  • 21.2.1 The nearest-neighbor field, and why exhaustive search is the bottleneck
  • 21.2.2 PatchMatch: randomized correspondence Barnes et al. 2009
  • 21.2.3 Applications: hole filling, retargeting, reshuffle
  • 21.2.4 Shift-Map image editing Pritch et al. 2009: editing as graph-cut labeling
• 21.3 Inpainting using millions of photographs
  • 21.3.1 Why self-similar inpainting isn't enough
  • 21.3.2 Scene completion from a huge database
  • 21.3.3 The data is the prior, and its modern opposite
• 21.4 Generative inpainting
  • 21.4.1 Filling a hole against saying what goes in it
  • 21.4.2 Inpainting in a latent space, and the seam it creates
  • 21.4.3 What it gets wrong
  • 21.4.4 Judging a fill with no reference
  • 21.4.5 An inpainted region is not evidence

▸ full collapsible outline of this part