Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ControlCity: A Multimodal Diffusion Model Based Approach for Accurate Geospatial Data Generation and Urban Morphology Analysis

Fangshuo Zhou, Huaxia Li, Rui Hu, Sensen Wu, Hailin Feng, Zhenhong Du, Liuchang Xu

TL;DR

Problem: heterogeneous VGI data quality limits accurate urban footprint generation and planning. Approach: ControlCity leverages a multimodal diffusion framework conditioned on text, road networks, land-use imagery, and metadata to transform open geospatial data into refined building footprints, evaluated across 22 cities with state-of-the-art metrics ($$FID=$50.94, $$MIoU=$0.36$$). Contributions: construction of an image-text-metadata-building footprint dataset, an enhanced ControlNet integration for multimodal conditioning, and demonstrated capabilities in urban morphology transfer, zero-shot city generation, and data-completeness assessment. Impact: enables more reliable urban planning analyses, morphology studies, and data-quality checks for VGI platforms, with potential extensions to 3D building modeling and vector-data preservation.

Abstract

Volunteer Geographic Information (VGI), with its rich variety, large volume, rapid updates, and diverse sources, has become a critical source of geospatial data. However, VGI data from platforms like OSM exhibit significant quality heterogeneity across different data types, particularly with urban building data. To address this, we propose a multi-source geographic data transformation solution, utilizing accessible and complete VGI data to assist in generating urban building footprint data. We also employ a multimodal data generation framework to improve accuracy. First, we introduce a pipeline for constructing an 'image-text-metadata-building footprint' dataset, primarily based on road network data and supplemented by other multimodal data. We then present ControlCity, a geographic data transformation method based on a multimodal diffusion model. This method first uses a pre-trained text-to-image model to align text, metadata, and building footprint data. An improved ControlNet further integrates road network and land-use imagery, producing refined building footprint data. Experiments across 22 global cities demonstrate that ControlCity successfully simulates real urban building patterns, achieving state-of-the-art performance. Specifically, our method achieves an average FID score of 50.94, reducing error by 71.01% compared to leading methods, and a MIoU score of 0.36, an improvement of 38.46%. Additionally, our model excels in tasks like urban morphology transfer, zero-shot city generation, and spatial data completeness assessment. In the zero-shot city task, our method accurately predicts and generates similar urban structures, demonstrating strong generalization. This study confirms the effectiveness of our approach in generating urban building footprint data and capturing complex city characteristics.

ControlCity: A Multimodal Diffusion Model Based Approach for Accurate Geospatial Data Generation and Urban Morphology Analysis

TL;DR

Problem: heterogeneous VGI data quality limits accurate urban footprint generation and planning. Approach: ControlCity leverages a multimodal diffusion framework conditioned on text, road networks, land-use imagery, and metadata to transform open geospatial data into refined building footprints, evaluated across 22 cities with state-of-the-art metrics (MIoU=$). Contributions: construction of an image-text-metadata-building footprint dataset, an enhanced ControlNet integration for multimodal conditioning, and demonstrated capabilities in urban morphology transfer, zero-shot city generation, and data-completeness assessment. Impact: enables more reliable urban planning analyses, morphology studies, and data-quality checks for VGI platforms, with potential extensions to 3D building modeling and vector-data preservation.

Abstract

Volunteer Geographic Information (VGI), with its rich variety, large volume, rapid updates, and diverse sources, has become a critical source of geospatial data. However, VGI data from platforms like OSM exhibit significant quality heterogeneity across different data types, particularly with urban building data. To address this, we propose a multi-source geographic data transformation solution, utilizing accessible and complete VGI data to assist in generating urban building footprint data. We also employ a multimodal data generation framework to improve accuracy. First, we introduce a pipeline for constructing an 'image-text-metadata-building footprint' dataset, primarily based on road network data and supplemented by other multimodal data. We then present ControlCity, a geographic data transformation method based on a multimodal diffusion model. This method first uses a pre-trained text-to-image model to align text, metadata, and building footprint data. An improved ControlNet further integrates road network and land-use imagery, producing refined building footprint data. Experiments across 22 global cities demonstrate that ControlCity successfully simulates real urban building patterns, achieving state-of-the-art performance. Specifically, our method achieves an average FID score of 50.94, reducing error by 71.01% compared to leading methods, and a MIoU score of 0.36, an improvement of 38.46%. Additionally, our model excels in tasks like urban morphology transfer, zero-shot city generation, and spatial data completeness assessment. In the zero-shot city task, our method accurately predicts and generates similar urban structures, demonstrating strong generalization. This study confirms the effectiveness of our approach in generating urban building footprint data and capturing complex city characteristics.
Paper Structure (21 sections, 6 equations, 12 figures, 8 tables)

This paper contains 21 sections, 6 equations, 12 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Diffusion Model
  4. Controllable Image Synthesis
  5. Urban building layout generation method
  6. Methodology
  7. Overview
  8. Dataset Construction
  9. Model Architecture
  10. Evaluation Metrics
  11. Experiment Setup
  12. Experiment and Results
  13. Experiment 1 - Analyze model performance
  14. Experiment 2 - Cross-region style transfer
  15. Experiment 3 - Zero-shot evaluation in unknown regions
  16. ...and 6 more sections

Figures (12)

  • Figure 1: Overall overview of ControlCity.
  • Figure 2: Data construction pipeline. Features are filtered from OSM data and combined with Wikipedia information to form text prompts using a LLM. The tile center coordinates are used as metadata. OSM building data is rasterized and paired with road network and land-use images. This process constructs a quadruple dataset of "image-text-metadata-building footprint."
  • Figure 3: The overall architecture of ControlCity. Road network and landuse image, text prompt and metadata are input into the model to generate building footprint image.
  • Figure 4: Comparison example of data generated by ControlCity and Pix2PixHD in 10 cities.
  • Figure 5: The composite results of generated data using Pix2PixHD and ControlCity methods. Examples from five cities are presented here.
  • ...and 7 more figures