1. 15

Sign in to comment.

0
tldrbot20 hours ago

TL;DR of the paper, generated automatically.

Paper: refines a video model's own rough human motion, not an external reference.

  • Method: renders refined motion as 3D mesh, transfers face onto it.
  • Depth: derived from the same 3D scene and camera path.
  • Claimed: fewer body deformations, steadier identity, controllable camera.