Reproduce “Unified Vector Floorplan Generation via Markup Representation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kaede Shiohara, Toshihiko Yamasaki
A unified framework named FMLM employs a novel Floorplan Markup Language (FML) to enable a single autoregressive transformer model to generate diverse vector floorplans. This approach handles various conditions and partial layouts, achieving competitive or superior performance compared to task-specific baselines across multiple generation tasks.
Key insights
- Representing complex spatial design problems, like floorplan generation, as a structured language enables a single generative model to perform a wide array of tasks previously requiring specialized architectures.
- Data augmentation through room permutation is crucial for autoregressive models to learn permutation-equivalent representations of unordered sets, such as rooms within a floorplan.
- Constrained decoding, guided by architectural heuristics, significantly improves the fidelity and plausibility of generated floorplans by preventing common errors in complex spatial arrangements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.