Urban-Sat
- U-Net
- semantic segmentation
- satellite imagery
Problem
Urban planners need road networks and building footprints pulled out of satellite imagery, which is slow and tedious to trace by hand.
What I built
A satellite-image segmentation application built on a U-Net model. It segments roads and building footprints from an input tile and uses those masks to generate layouts.
Result
Used to identify road networks from satellite images and to generate house layouts for urban-planning workflows.