Mapping the walk: A scalable computer vision approach for generating sidewalk network datasets from aerial imagery

Maryam Hosseini, Andres Sevtsuk, Fabio Miranda, Roberto M. Cesar, Cláudio T. Silva

Abstract: Highlights•Introduces Tile2Net, an end-to-end open-source framework for creating georeferenced pedestrian networks from aerial imagery.•Implements the framework on Boston, Cambridge, New York City and Washington with a relatively high degree of accuracy.•Demonstrates a novel technique for creating semantic segmentation annotation masks from available planimetric GIS data.•The tool, open-source input data links and model weights and are made publicly available.