WalkGeoAI: Street-Level Pedestrian Density Dataset (20 Global Cities)

Overview This repository contains the official dataset for the study: "WalkGeoAI: An open-source framework for estimating street-level pedestrian density using globally available open data." It provides high-resolution, street-segment-level pedestrian density estimates and comprehensive built-environment feature matrices for 20 major global cities. The dataset was generated using the open-source WalkGeoAI framework, which leverages a spatially-aware Graph Neural Network (GatedGCN) and multi-source global open data to perform zero-shot spatial transfer inference. This dataset overcomes the traditional reliance on expensive local sensing infrastructure, offering a standardized baseline of pedestrian activity across diverse urban morphologies worldwide. Dataset Structure The dataset is organized into 20 individual compressed archives (.zip), each representing a specific city. Upon extracting a city's archive, you will find the following standardized structure: 📄 road_flows.geojson The core spatial vector file containing the city's street network geometries and the final model predictions. Standard Fields: id, pred_density (estimated relative pedestrian density, normalized 0-1), pred_uncertainty (spatial uncertainty calculated via Monte Carlo Dropout), and geometry. 🌟 Special Note for Beijing & New York: For these two cities, the file additionally contains an actual_density field. This represents the ground-truth pedestrian activity data used to validate the model, making it highly valuable for benchmark testing. 📁 data file/ Contains 7 tabular data files (.csv) storing the 204 multi-scale built-environment features extracted for every street segment (linked via the id field). The files are: net_topology_features.csv: Street network topology, centrality, and intersection metrics. pop_demand_features.csv: Local population densities and spatial gradients. terrain_topography_features.csv: Street slopes and relative elevations. urban_context_features.csv: Macroscopic urban structure (distances to centers/edges). bldg_morphology_features.csv: Building footprints, FAR, and street-wall ratios. landcover_greenness_features.csv: NDVI stats and WorldCover environmental class proportions. poi_density_buf200.csv: Granular semantic point-of-interest (POI) densities. 📁 shp file/ Contains the raw and intermediate GIS files used to calculate the features in the pipeline. Typical files include: Administrative_buildup.geojson: The defined urban/administrative boundary. 3DGloBFP.geojson: Processed building footprints. POI.csv: Standardized point-of-interest dataset. pop_clipped.tif: Clipped WorldPop population raster. NASADEM.tif: Topographic elevation raster. WorldCover.tif: ESA land cover classification raster. Code Availability & Feature Dictionary The complete Python pipeline used to extract these features and generate the predictions, along with the detailed Feature Dictionary explaining all variables, is fully open-sourced. GitHub Repository: https://github.com/UrbanMobility-y/WalkGeoAI License This dataset is made available under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.