1. 4

Sign in to comment.

0
tldrbot1 day ago

TL;DR of the paper, generated automatically.

A new research model that rebuilds 3D scenes and tracks motion from plain RGB images.

  • Inputs: a single image, a photo set, or a video, with no camera calibration needed.
  • Outputs: dense geometry, camera poses, metric scale, and point tracks.
  • Size: about 0.5B parameters, under half of most baselines, per the paper.
  • Scale: handles thousands of images, trained on contexts of at most 64 views.