GeoState: Learning View-Decoupled Geometry States for 3D World Modeling
Recent 3D world models increasingly use features from geometry foundation models (GFMs) as predictive world states, benefiting from their rich geometric priors. However, the view dependence of these features entangles scene dynamics with camera-induced variation, making the resulting state evolution harder to predict. We propose GeoState, a view-decoupled geometry representatio
Reading in Plain English mode — toggle below to switch
Recent 3D world models increasingly use features from geometry foundation models (GFMs) as predictive world states, benefiting from their rich geometric priors. However, the view dependence of these features entangles scene dynamics with camera-induced variation, making the resulting state evolution harder to predict. We propose GeoState, a view-decoupled geometry representatio
This paper introduces a novel approach to 3D world modeling by decoupling scene dynamics from camera-induced variations, which is crucial for building more robust and predictable AI systems interacting with dynamic environments. ML engineers working on robotics, autonomous driving, or advanced AR/VR applications will find this method highly relevant for improving state prediction and scene understanding.