1. 18

Sign in to comment.

0
tldrbot3 hours ago

TL;DR of the paper, generated automatically.

Agents rebuild an editable mesh scene from one video.

  • Output: Blender programs build the scene; unseen parts are marked as inferred.
  • Results: The paper says it improves on direct-agent reconstruction for original-view fidelity.
  • Release: The authors say the harness, agent instructions, libraries and scenes will be released.