FloorplanMAE:A self-supervised framework for complete floorplan generation from partial inputs (2506.08363v1)

Abstract: In the architectural design process, floorplan design is often a dynamic and iterative process. Architects progressively draw various parts of the floorplan according to their ideas and requirements, continuously adjusting and refining throughout the design process. Therefore, the ability to predict a complete floorplan from a partial one holds significant value in the design process. Such prediction can help architects quickly generate preliminary designs, improve design efficiency, and reduce the workload associated with repeated modifications. To address this need, we propose FloorplanMAE, a self-supervised learning framework for restoring incomplete floor plans into complete ones. First, we developed a floor plan reconstruction dataset, FloorplanNet, specifically trained on architectural floor plans. Secondly, we propose a floor plan reconstruction method based on Masked Autoencoders (MAE), which reconstructs missing parts by masking sections of the floor plan and training a lightweight Vision Transformer (ViT). We evaluated the reconstruction accuracy of FloorplanMAE and compared it with state-of-the-art benchmarks. Additionally, we validated the model using real sketches from the early stages of architectural design. Experimental results show that the FloorplanMAE model can generate high-quality complete floor plans from incomplete partial plans. This framework provides a scalable solution for floor plan generation, with broad application prospects.

• The paper introduces FloorplanMAE, a novel self-supervised model that reconstructs complete floorplans from sparse input data.
• It leverages Masked Autoencoders and Vision Transformers to outperform baselines under diverse masking strategies with high fidelity metrics.
• The framework facilitates rapid architectural prototyping by automating full floorplan generation from fragmented designs.

Analysis of "FloorplanMAE: A Self-Supervised Framework for Complete Floorplan Generation from Partial Inputs"

The paper entitled "FloorplanMAE: A Self-Supervised Framework for Complete Floorplan Generation from Partial Inputs" presents a novel approach in the field of architectural design, specifically focusing on the automation of floorplan reconstruction from partial data inputs. The methodology leverages self-supervised learning, utilizing Masked Autoencoders (MAE) integrated with Vision Transformers (ViT), marking an intersection of advanced AI techniques with architectural innovation.

Key Contributions and Methodology

FloorplanMAE is introduced as a framework that intelligently reconstructs incomplete architectural floor plans into full layouts. The authors constructed a dedicated dataset, FloorplanNet, which comprises a rich repository of architectural floor plans to train and validate their model. The FloorplanMAE model is composed of a lightweight encoder and an asymmetric decoder. The encoder operates selectively on visible patches of the floor plan, embedding positional data and processing through Transformer blocks, while the decoder incorporates masked tokens for learning, facilitating reconstruction with sparse information inputs.

Experimental Results

The paper thoroughly evaluates the reconstruction performance of FloorplanMAE across a variety of masking strategies: random, center, edge, perimeter, one-sided, and corner masking. The experimental results are illustrated with both qualitative and quantitative assessments. Notably, FloorplanMAE demonstrates a strong capacity to maintain functional zoning and structural integrity, despite diverse masking conditions. The model exhibits superior results compared to baseline methodologies like pix2pix and cycleGAN, showing high fidelity indicated by metrics such as FID, PSNR, and SSIM.

The effectiveness of FloorplanMAE is particularly evident in scenarios with high masking ratios such as random masking with 80% coverage, where it preserves the overall layout and functional zoning despite deviations in specific details. The varied strategies underscore the adaptability of FloorplanMAE in understanding and predicting floorplan structures under incomplete data conditions.

Implications and Future Work

FloorplanMAE introduces significant implications for the architectural design process. By enabling the extraction of complete floorplans from fragmented inputs, it proposes a new paradigm that aids architects in rapid prototyping and design ideation, potentially leading to substantial time-saving and efficiency improvements in iterative floorplan development.

Moreover, the framework posits questions regarding the balance between creativity and control within automated design processes. Specifically, the authors invite discourse on whether traditional metrics adequately capture the rationality and practical applicability of generated architectural designs. This opens avenues for future exploration in refining AI-driven architectural design methodologies, particularly in enhancing the interpretability and meaningfulness of generated outputs within real-world applications.

Conclusion

"FloorplanMAE: A Self-Supervised Framework for Complete Floorplan Generation from Partial Inputs" offers substantial advancements in the computational generation of architectural designs. Utilizing self-supervised learning techniques, it paves the way for more integrated, automatic processes in floorplan generation, addressing the challenges of structured design logic and geometric abstraction. The promising results and open questions presented in the paper lay a strong foundation for continued exploration and refinement in AI-assisted architectural design, pushing the boundaries of how AI can leverage complex datasets to deliver practical and insightful design solutions.