Elevating the ambitions of TDEI and OpenSidewalks, the Annotations for Pedestrian Environment (APE) dataset addresses a critical urban mobility challenge: accurately mapping pedestrian pathways, including sidewalks and crossings. This project introduces a novel training dataset featuring aerial satellite imagery, street map imagery, and precise annotations of sidewalks, crossings, and corner bulbs across various cities. The APE dataset is crucial for a broad spectrum of applications, enhancing pedestrian pathway assessments, and fostering improvements in multi-modal navigation and urban planning.
APE Dataset Overview
The dataset spans $2,700 km2 land area, covering select regions from different cities. It can be used for various learning tasks related to segmenting and understanding pedestrian environments. The APE dataset encompasses aerial imagery, street map data, and human annotation from select regions across the Americas:
- United States: Los Angeles, CA; Bellevue, WA; Seattle, WA
- View the Los Angeles APE dataset ↗; If you are having trouble accessing the dataset, please reach out to yz325@uw.edu
- Ecuador: Quito
- Brazil: Sao Paulo
- Chile: Santiago, Gran Valparaiso
Dataset Composition
Each dataset sample includes:
- Aerial Imagery (xxxx_sat_train.png): Satellite images capturing the top-down view along major roads and intersections within each region.
- Rasterized Street Map Tiles (xxxx_road_train.png): Rasterized representations of street maps aligned with the aerial imagery.
- Rasterized Annotations for Pedestrian Pathways (xxxx_gt_mask.png): Human-validated segmentation masks for pedestrian pathways, supporting semantic segmentation tasks with classes including sidewalks, crossings, and corner bulbs.
- Color-Coded Pathway Visualizations (xxxx_gt_color.png): Color-coded segmentation masks for easier interpretation.
Alignment and Registration
Each sample set is precisely aligned and co-registered within the same geographic bounding box, ensuring consistent spatial referencing across images and maps.
Additional Resources
Read more about APE: An Open and Shared Annotated Dataset for Learning Urban Pedestrian Path Networks ↗
For extended analyses, see PathwayBench for co-registered road graphs and pedestrian pathway graphs available in GeoJSON format.