Event
Urbex: An Intelligent Agent Spatial Localization Framework for City-Scale 3D Scenes
1 reports · 1 sources
The research team published the Urbex paper, a vision-language framework for agents in city-scale 3D scenes. It shifts from static ranking that relies on pre-generated candidate boxes to active spatial search, progressively narrowing the range through three tools—landmark querying, local zoom-in, and oblique views—and uses GRPO reinforcement learning to optimize the tool-invocation policy. The paper states that Urbex achieves competitive localization accuracy on the CityRefer benchmark; the model was trained only on CityRefer, and zero-shot experiments directly applied to STPLS3D-Refer also show good cross-scene generalization ability.
Report timeline
- NeurIPS 2026 Paper Urbex: Agent-Driven City-Scale 3D Spatial Localization
Urbex is an agentic vision-language framework for city-scale 3D scenes that shifts from static ranking based on pre-generated candidate boxes to active spatial search, progressively narrowing the search area through three tools—landmark query, local zoom-in, and oblique view—and uses GRPO reinforcement learning to optimize tool-invocation strategies.
CVer · 微信公众号 ·