FloorFlex Personalized Floor Plan Recommendation Through Deep Learning and GAN

Abstract

"Traditional floor planning processes are often marred by complexity, lack of customization, and the potential for costly mistakes. Current floor plan recommendations typically offer minimal requirements and require more user-centric designs. They often need to align with new floor plan trends and use similar data for all recommendations, disconnecting the architect and the user. Users may need help to express their needs clearly, and architects may find it challenging to design plans that meet those requirements, leading to significant time wastage for both parties. To solve this problem, the author extensively studied existing challenges in floor plan design and implemented Deep Learning and GAN technologies into the software's architecture. The system features a user-friendly interface that allows users to input their preferences directly. Based on these inputs, the GAN-driven recommendation engine then refines and personalizes floor plans. The technical approach involved designing a multi-layer Convolutional Neural Network within the GAN framework to accurately model and predict spatial arrangements, ensuring the generated plans are functional and aesthetically pleasing. Based on the evaluation results from the Tell2Design dataset with CGAN, this model achieved an Inception Score of 1.00 ± 2.0 and a Frechet Inception Distance of approximately 1600, even though it was trained with a small dataset. These metrics indicate a high level of diversity and quality in the generated floor plans despite the challenges posed by the dataset size. The IS score suggests good perceptual quality, while the FID score reflects reasonable similarity to real floor plans in the dataset. This performance underscores the effectiveness and reliability of the CGAN-based approach in generating accurate and user-preferred floor plans, validating its potential to enhance architectural and interior design practices through advanced AI technologies."