I-Design: Personalized LLM Interior Designer

I-Design: Personalized LLM Interior Designer

연구 분야: Databases

학회: European Conference on Computer Vision

Interior design allows us to express our identities and live according to our preferences, with each design reflecting our unique personalities. However, it is not easy for non-professionals to realize these visions, as it requires aligning functional and aesthetic expectations with the constraints of physical and digital spaces. This complexity often renders interior design a luxury. To make interior design more accessible, we present I-Design, a personalized interior design tool that enables users to generate and visualize their design goals through natural language communication. I-Design utilizes a team of large language model agents that engage in dialogue and logical reasoning to transform textual user input into feasible scene graph designs with relative object relationships. Following this, an effective placement algorithm determines the optimal location for each object within the scene. The final design is then constructed in 3D by retrieving and integrating assets from an existing object database. Additionally, we propose a new evaluation protocol that employs a vision-language model and complements the design pipeline. Extensive experiments demonstrate that I-Design outperforms existing methods in delivering high-quality 3D design solutions and aligning with abstract concepts that match user input, showcasing its advantages in both detailed 3D arrangement and conceptual fidelity.