Skip to main content

Vid2Avatar, CVPR 2023

· One min read
SeulGi Hong

Today, I'm going to review the paper for this CVPR 2023, Vid2Avatar.

"Vid2Avatar: 3D Avatar Reconstruction from Videos in the Wild via Self-supervised Scene Decomposition"​

Summary

image

  • casting the segmentation and surface as a joint 3d optimization problem
    • external segmentation model X
  • spherical space -> sample points are divided into in and out of the sphere
  • inner space points -> canonical SDF -> resampling the points by making use of surface information -> canonical texture -> render RGB
  • outer space points (randomly sampled) -> SDF (static scene, background) -> combined with dynamic foreground(human)