Polygonal Building Segmentation by Frame Field Learning

TL;DR

This paper introduces a deep learning approach that predicts frame fields alongside segmentation masks to improve vector polygon extraction of buildings from remote sensing images, bridging raster outputs and GIS application needs.

Contribution

It presents a novel multi-task neural network that predicts frame fields for better polygonization of building footprints in remote sensing imagery.

Findings

Enhanced segmentation quality through multi-task learning

Effective polygonization using predicted frame fields

Open-source code available for reproducibility

Abstract

While state of the art image segmentation models typically output segmentations in raster format, applications in geographic information systems often require vector polygons. To help bridge the gap between deep network output and the format used in downstream tasks, we add a frame field output to a deep segmentation model for extracting buildings from remote sensing images. We train a deep neural network that aligns a predicted frame field to ground truth contours. This additional objective improves segmentation quality by leveraging multi-task learning and provides structural information that later facilitates polygonization; we also introduce a polygonization algorithm that utilizes the frame field along with the raster segmentation. Our code is available at https://github.com/Lydorn/Polygonization-by-Frame-Field-Learning.