Conditional room layout generation based on graph neural networks

Zhihan Yao, Yuhang Chen, Jiahao Cui, Shoulong Zhang, Shuai Li, Aimin Hao

Abstract: Highlights•We propose an end-to-end model for 3D indoor scene synthesis using graph latents.•Unlike floor plans, our method uses graph-based wall nodes for stronger constraints.•Newly latent representations provide function and semantic guidance for optimization.