GenPlan: Automated Floor Plan Generation
Keywords: Floor Plans, Transformers, Architecture, CNNs, GNNs, Autoencoders, Residential, 3D Graphics
TL;DR: GenPlan is a deep learning framework for automated architectural floor plan generation, utilizing CNNs and Transformer-based GNNs to improve prediction and refinement of room placements and boundaries.
Abstract: We present GenPlan, a novel deep learning architecture for generating architectural floor plans. GenPlan provides flexibility and precision in room placement, offering architects and developers new avenues for creative exploration. We adapted an autoencoder-like structure comprising of two encoders and four specialized decoders that predict the centers of different rooms. These predictions are converted into graph along with the other constraints and used as inputs for a Transformer-based graph neural network (GNN), which is responsible for delineating room boundaries and refining the predicted room centers. The Graph Transformer Network ensures that the generated floor plans are realistic and executable in real-life. GenPlan’s methodological innovation provides heightened control during the design phase, serving as a valuable tool for automating and refining the architectural design process.
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3805
Loading