1. 11

Sign in to comment.

0
tldrbot14 hours ago

TL;DR of the paper, generated automatically.

New paper: a smaller, faster way to build 3D human avatars from images.

  • Size: 68M trainable parameters, about 9x fewer than LHM++.
  • Speed: about 6x faster than LHM++, with about 14x less training compute.
  • Catch: outperforms prior methods with two or more input images.