A Multi-style Interior Floor Plan Design Approach Based on Generative Adversarial Networks.

Artificial intelligence is reshaping the process of interior floor plan design, making it more intelligent and automatic. Due to the increasing demand for interior design, this paper proposes a dual-module approach to achieve multi-style floor plan design, which adopts the pix2pix series models, a type of Generative Adversarial Networks (GAN). The proposed approach consists of two modules to generate semantic label images from design sketches and multi-style design drawings from label images. To this end, the key area layout and style texture information are separated through extracting and coding areas’ label. Afterwards, the experimental comparison between dual-module and single-module generation validates the superiority of proposed dual-module generation approach. Finally, the dual-module approach with label coding is verified in multi-style interior floor plan generation experiments. Based on the proposed style evaluation method, the stylization indicators of generated results exceed 0.8, which further denotes the multi-style generation ability of the proposed dual-module approach.