In the GIS field, especially when creating CIM digital twin scenarios, building white model data is often required. I have previously used many such building white model datasets (such as global building footprint data released by Google and Microsoft), but I frequently encountered situations where only the "footprint" was available without height information, or the height was randomly generated. A few days ago, I came across an interesting dataset online: GlobalBuildingAtlas, which can precisely solve the problem of missing height information. The data is global in scope, so interested friends should not miss it.

Introduction to the GlobalBuildingAtlas Dataset

GlobalBuildingAtlas is a global, complete, and open-source building dataset released by the Technical University of Munich. Unlike previous datasets, GlobalBuildingAtlas not only provides 2D polygonal footprints of buildings but also innovatively provides building heights and LoD1 level 3D models. The currently released models contain height information for approximately 2.75 billion buildings. All data is extracted from satellite imagery.

Core Highlights

  1. Global Coverage, Complete Data: Aims to provide building data on a global scale, filling gaps in existing datasets for certain regions.
  2. 3D Data: Integrates building height information, directly generating LoD1 3D models, significantly lowering the barrier to modeling from 2D data to 3D scenes.
  3. Open Source: Not only provides data but also open-sources the complete code from satellite image processing to final 3D model generation.

The current data includes the following:

Data Download and Usage

Currently, the official provides several methods for using the data.

WFS Service

If you use QGIS or ArcGIS, you can directly load the WFS service to view and analyze the data online without downloading the massive dataset. WFS service address:

https://tubvsig-so2sat-vm1.srv.mwn.de/geoserver/ows?

PS: Access is currently a bit slow; I couldn't load it locally.

Online Viewer

The project also provides an online viewer address:

https://tubvsig-so2sat-vm1.srv.mwn.de/

It opens as follows:

Download Data Locally

Data download address:

https://mediatum.ub.tum.de/1782307

Total size: 8 GB, 8 files.

Core Algorithms

If you not only want to use the data but also wish to further tune the extraction algorithms, you can check the official open-source code:

https://github.com/zhu-xlab/GlobalBuildingAtlas?tab=readme-ov-file

Several main modules:

  • im2bf: Generates building polygons from satellite images.
  • im2bh: Monocular building height estimation.
  • fuse_bf: Fusion and optimization of multi-source data.
  • make_lod1: LoD1 model generation.

Summary

Based on current feedback I've seen, some extractions are not particularly accurate. I have also compared several famous buildings, and this is indeed the case. However, it is much more useful than purely 2D data or data with random heights. If your project happens to need this kind of background data, don't miss it—download it and give it a try.