Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
RoofDiT introduces a novel generative framework for synthesizing and reconstructing 2D roof graphs by modeling them as vertex-edge graphs, moving away from traditional geometric rules. The framework employs a two-stage design, utilizing a diffusion transformer for vertex generation and an edge prediction module for graph topology inference, while enhancing geometric fidelity through geometry-aware attention and conditioning on aerial images. Experimental results indicate that RoofDiT outperforms existing methods in graph generation quality, particularly excelling in footprint-conditioned synthesis and image-guided reconstruction tasks.
RoofDiT achieves state-of-the-art edge F1 scores for image-guided roof reconstruction by leveraging a novel diffusion transformer approach that directly models roof structures as graphs.
We present RoofDiT, a generative framework for 2D roof graph synthesis and reconstruction. Roofs are compactly described as planar graphs of junctions and structural edges, but existing methods often rely on fixed geometric rules or direct reconstruction objectives. RoofDiT instead models roof structures directly as vertex-edge graphs and learns a conditional generative prior over their geometry and connectivity. Our framework follows a two-stage design: a diffusion transformer generates roof vertices, and an edge prediction module infers the corresponding graph topology. To improve geometric fidelity, RoofDiT combines relative geometry-aware attention with footprint and aerial-image conditioning, while using an alignment regularizer to encourage common horizontal, vertical, and diagonal roof patterns. The same model supports unconditional generation, footprint-conditioned synthesis, and image-guided reconstruction by changing the conditioning signal. Experiments show improved graph generation quality over a diffusion baseline, favorable performance against a straight-skeleton prior in the footprint-conditioned setting, and the highest edge F1 among compared methods for image-guided reconstruction.