Parameterized Interior Floor Plan Design Based on Differentiable Renderer

Abstract

AbstractCreating floor plans for indoor scenes is a fundamental task of architecture and interior design. Since a high-quality floor plan requires proper space division and location distribution for rooms inside the scene, such a task always require professional designers to spend huge efforts and workloads. In this paper, we intend to leverage the artificial intelligence method to relieve the manual work of interior floor plan design. Specifically, we employ the differentiable renderer, which is an optimizer that adjusts the parameters of certainly given mesh primitives. The optimization process is under the constraints of the input outer boundary, overlap, and weights, to ensure the design is parameterized, and suitable for interior floor plans with plausible space division. We also provide post-processing to refine the optimized mesh primitives and convert them to an interior floor plan. We conduct both ablation and controlled experiments to verify the effectiveness of our method in floor plan design.KeywordsComputer-aided designFloor plan synthesisDifferentiable renderer