CityCode: Building Agent-Operable 3D Cities from Sparse Aerial Observations

Key points

• Urban spatial intelligence calls for 3D cities that agents can measure, interpret, and modify across successive tasks.1
• We therefore formulate a new setting for grounded urban world construction: generating large-scale, instance-level parameterized 3D cities from one low-altitude aerial RGB image per region and known camera parameters, while recovering metrically scaled buildings and explicit road geometry and connectivity.1
• Experiments on CitySample-Aerial and UrbanScene3D demonstrate substantially more accurate metric structure recovery and superior visual fidelity to the target cities compared with the evaluated baselines.1
• Together, CityCode and Air2Code turn sparse aerial observations into persistent, agent-operable urban worlds for long-horizon urban spatial intelligence.1

Sources (1)

• [1]CityCode: Building Agent-Operable 3D Cities from Sparse Aerial Observations
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 01:14 PM

  We introduce CityCode, an agent-operable representation that unifies metric geometry, urban topology, visual semantics, and 3D assets through persistent entity identities, with dependency-aware versioned execution for validated updates and historical restoration.

  Urban spatial intelligence calls for 3D cities that agents can measure, interpret, and modify across successive tasks.

Extractive summary: sentences quoted from the sources.