Reproduce “HouseDiffusion: Vector Floorplan Generation via a Diffusion Model with Discrete and Continuous Denoising” 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
Mohammad Amin Shabani, Sepidehsadat Hosseini, Yasutaka Furukawa
HouseDiffusion introduces a diffusion model for directly generating vector-graphics floorplans, addressing the limitations of raster-based methods by incorporating a novel dual discrete and continuous denoising process. This approach significantly improves diversity and compatibility with input constraints, producing geometrically precise and architecturally sound floorplans that include non-Manhattan designs and user-controlled room shapes.
Key insights
- A dual denoising strategy, combining continuous regression for smooth approximations and discrete coordinate inference for exact geometric alignment, is crucial for generating precise vector graphics.
- A Transformer-based architecture with structured attention (component-wise, global, and relational cross-attention) is essential for reasoning about local room shapes, global layout coherence, and adherence to input graph constraints.
- Vector-based representation directly addresses the limitations of raster-based methods, enabling precise control over room shapes, non-Manhattan layouts, and seamless integration with CAD software.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.