VLM-based automatic multi-granularity graph representation of building layouts for design informatics

Computer Science > Artificial Intelligence
[Submitted on 8 May 2026]
Title:VLM-based automatic multi-granularity graph representation of building layouts for design informatics
View PDF HTML (experimental)Abstract:Architectural floorplan images encode rich relational knowledge among functional spaces, which underpins design retrieval, knowledge-based reasoning, and BIM enrichment through the building lifecycle. However, it remains challenging to automatically construct task-adaptive graph representations for public buildings. To address this gap, we first define a multi-granularity Level-of-Graphs (LoGs) for public building layouts. Methodologically, we present a Vision-Language Model (VLM)-based automatic LoG construction through node identification, edge inference, text parsing, and graph coarsening. VLM-generated representations are systematically evaluated and tested in real-world tasks, using 147 academic library floorplans worldwide as a case study. Experiments showed VLM-generated graphs were broadly consistent with human-labeled graphs (matched node ratio >= 92%; 509.3 s per floor plan for three-LoG graph generation). Meso-grained graphs yield the best node-level zone prediction (Macro F1 = 0.647, at 65% of fine-grained complexity), while coarse-grained graphs are most effective for graph-level layout quality evaluation (Spearman's \r{ho} = 0.610, at 16% of fine-grained complexity). By enabling scalable, annotation-free extraction of structured layout information from floorplan images, this study advances design informatics by converting plan images into knowledge representations, thereby enhancing the utilization of design information across the building life cycle.
References & Citations
Loading...
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Facts Only

* A Vision-Language Model is used to automatically construct multi-granularity Level-of-Graphs (LoGs) for public building layouts.
* The methodology includes node identification, edge inference, text parsing, and graph coarsening.
* Evaluation used 147 academic library floorplans worldwide as a case study.
* VLM-generated graphs showed a node ratio $\ge$ 92% consistency with human-labeled graphs.
* Generating three-LoG graphs took 509.3 seconds per floor plan.
* Meso-grained graphs optimized node-level zone prediction (Macro F1 = 0.647 at 65% complexity).
* Coarse-grained graphs optimized graph-level layout quality evaluation (Spearman's $\rho$ = 0.610 at 16% complexity).

Executive Summary

A Vision-Language Model is used to automatically construct multi-granularity Level-of-Graphs (LoGs) from architectural floorplan images. The process involves node identification, edge inference, text parsing, and graph coarsening. These VLM-generated representations were evaluated using 147 academic library floorplans worldwide as a case study. Experiments found that the generated graphs showed broad consistency with human-labeled graphs, achieving a node ratio greater than or equal to 92%. For generating three-LoG graphs, the process required 509.3 seconds per floor plan. Meso-grained graphs provided the best performance for node-level zone prediction, yielding a Macro F1 score of 0.647 at 65% of fine-grained complexity. Coarse-grained graphs were most effective when evaluating graph-level layout quality, achieving a Spearman's rho of 0.610 at 16% of fine-grained complexity. The study concludes that this method advances design informatics by converting plan images into knowledge representations through annotation-free extraction.

Full Take

The work demonstrates a pathway toward automated knowledge extraction from complex visual data, positioning VLM technology as a tool for bridging the gap between static architectural plans and dynamic knowledge graphs relevant to design informatics. The observed performance metrics suggest that tailoring the graph granularity—choosing between meso- or coarse-grained representations—is critical for application-specific utility; no single representation is universally superior across all tasks. This signals an important pattern: abstracting visual information into structured, relational knowledge structures is feasible, provided the complexity of the output is systematically controlled. The implication for design informatics is significant: removing manual annotation from plan interpretation could unlock scalable reasoning and enrichment across the building lifecycle. However, the results focus on a specific context (library floorplans) and specific evaluation metrics; future work must address how these VLM-derived graphs perform when dealing with non-standard architectural notations or highly complex, multi-level functional relationships found in commercial or public buildings outside this test set. What assumptions are made about the universal interpretability of the learned relational structures across varied spatial scales? How does the consistency observed hold up when human experts impose different conceptual boundaries during labeling?

Sentinel — Human

Confidence

This text reads like a summary of a technical academic paper, characterized by precise terminology and quantitative results, suggesting human authorship in a research context.

Signals Detected

Moderate sentence length variance; academic tone is consistent with research reporting.

High flow and focus directly addressing a technical problem and proposed solution without excessive hedging.

Structure aligns perfectly with a typical research abstract/summary format, citing specific metrics (F1, rho).

Claims regarding methodology and reported experimental results appear concrete and internally consistent.

Human Indicators

The specific statistical results (Macro F1 = 0.647, Spearman's rho = 0.610) and the detailed breakdown of VLM-based steps suggest original research reporting rather than generic synthesis.