UrbanIR: Large-Scale Urban Scene Inverse Rendering from a Single Video
This addresses the challenge of generating photorealistic and editable renderings from sparse video data for applications like urban planning and simulation, representing a strong domain-specific advancement.
The paper tackles the problem of realistic free-viewpoint rendering of urban scenes under various lighting conditions from a single video, achieving significant improvements over state-of-the-art methods by accurately inferring shape, albedo, visibility, and illumination.
We present UrbanIR (Urban Scene Inverse Rendering), a new inverse graphics model that enables realistic, free-viewpoint renderings of scenes under various lighting conditions with a single video. It accurately infers shape, albedo, visibility, and sun and sky illumination from wide-baseline videos, such as those from car-mounted cameras, differing from NeRF's dense view settings. In this context, standard methods often yield subpar geometry and material estimates, such as inaccurate roof representations and numerous 'floaters'. UrbanIR addresses these issues with novel losses that reduce errors in inverse graphics inference and rendering artifacts. Its techniques allow for precise shadow volume estimation in the original scene. The model's outputs support controllable editing, enabling photorealistic free-viewpoint renderings of night simulations, relit scenes, and inserted objects, marking a significant improvement over existing state-of-the-art methods.