Urbex：面向城市级3D场景的智能体空间定位
原标题：Paper page - Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes
AI 摘要
该论文提出 Urbex，一个面向城市级 3D 场景的智能体视觉语言框架，将空间定位从候选排序改为主动空间搜索。VLM 智能体可查询地标、缩放局部区域、渲染斜视图并输出定位框，并通过在 CityRefer 上的强化学习优化工具使用策略。实验显示 Urbex 在 CityRefer 上取得最佳 Hit@1 和 IoU，无需候选框且速度显著快于最强排序基线，在 STPLS3D-Refer 上的零样本结果也表明跨场景鲁棒性提升。代码、模型、数据和 Colab 演示均已公开。
正文节选
Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes Abstract City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes. We propose Urbex, an agentic vision-language framework that instead casts grounding as active spatial search. Urbex represents a city-scale 3D scene as an interactive multi-view environment, allowing a VLM agent to query landmarks, zoom into local regions, render obliq