GeoState: Learning View-Decoupled Geometry States for 3D World Modeling

Key points

• Recent 3D world models increasingly use features from geometry foundation models (GFMs) as predictive world states, benefiting from their rich geometric priors.1
• However, the view dependence of these features entangles scene dynamics with camera-induced variation, making the resulting state evolution harder to predict.1
• Specifically, a scene tokenizer aggregates multi-view GFM features into a view-decoupled scene state, which a geometry decoder queries using camera rays.1
• Experiments show that GeoState outperforms prior 3D world models on Waymo, supports geometrically consistent cross-view querying, and generalizes zero-shot to KITTI and nuScenes.1

Sources (2)

• [1]GeoState: Learning View-Decoupled Geometry States for 3D World Modeling
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 05:56 PM

  We propose GeoState, a view-decoupled geometry representation for predictive 3D world modeling that separates scene-state evolution from camera-dependent observation.

  Recent 3D world models increasingly use features from geometry foundation models (GFMs) as predictive world states, benefiting from their rich geometric priors.

• [2]GeoState: Learning View-Decoupled Geometry States for 3D World Modeling
  Hugging Face Daily Papers · Oct 15, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.