Abstract
The world is covered with millions of buildings, and precisely knowing eachinstance's position and extents is vital to a multitude of applications.Recently, automated building footprint segmentation models have shown superiordetection accuracy thanks to the usage of Convolutional Neural Networks (CNN).However, even the latest evolutions struggle to precisely delineating borders,which often leads to geometric distortions and inadvertent fusion of adjacentbuilding instances. We propose to overcome this issue by exploiting thedistinct geometric properties of buildings. To this end, we present DeepStructured Active Contours (DSAC), a novel framework that integrates priors andconstraints into the segmentation process, such as continuous boundaries,smooth edges, and sharp corners. To do so, DSAC employs Active Contour Models(ACM), a family of constraint- and prior-based polygonal models. We learn ACMparameterizations per instance using a CNN, and show how to incorporate allcomponents in a structured output model, making DSAC trainable end-to-end. Weevaluate DSAC on three challenging building instance segmentation datasets,where it compares favorably against state-of-the-art. Code will be madeavailable.