LightCrafter: A Structured Framework for Function-Aware Indoor Lighting Design.

Indoor lighting design involves complex spatial, functional, and aesthetic considerations, making it challenging to achieve satisfactory results without expertise. Poor lighting distribution-such as excessive glare or uneven illumination-misalignment with scene-specific functional requirements like over- or under-illumination, and inefficient control systems significantly undermine both energy efficiency and lighting quality. In this work, we come up with a structured framework for function-aware indoor lighting design that formulates lighting design as a structured decision-making problem. Considering the scene as a hierarchical scene graph that captures the spatial and semantic relationships between scene elements, we use a pre-trained LLM to reconstruct the scene into a lighting structure graph under the guidance of a lighting knowledge base, containing nodes with lighting requirements and possible lighting nodes. Subsequently, in the lighting simulation engine, reinforcement learning is used to optimize the parameter values of the lighting nodes in the lighting structure graph to meet the lighting design requirements. Extensive quantitative and qualitative experiments show that our method outperforms existing methods for automatic indoor lighting design. It not only generates high-quality lighting designs, but also perceives the functions of the scene, allowing users to easily achieve functionally controllable indoor lighting designs through structured scene editing.