📹Can 3D scenes be encoded into and rendered from a set of compressed tokens? New work "SceneTok" from the Geometric Representation Learning group at MPI for Informatics shows that they can, and that this approach pairs naturally with generative rendering to handle uncertainty in observations.

↪️More information on the project page: https://mpi-inf.link/scenetok

Accepted at CVPR2026

#research #computerscience #paper

SceneTok

SceneTok: A Compressed, Diffusable Token Space for 3D Scenes.