Regularization of Building Boundaries in Satellite Images using Adversarial and Regularized Losses
This addresses the need for regularized building footprints in satellite imagery applications, though it is incremental as it builds upon existing segmentation models.
The paper tackles the problem of irregular building boundaries in satellite images by introducing a method that refines and regularizes them using a fully convolutional neural network with adversarial and regularized losses, achieving equivalent accuracy and completeness to Mask R-CNN while producing more visually pleasing boundaries.
In this paper we present a method for building boundary refinement and regularization in satellite images using a fully convolutional neural network trained with a combination of adversarial and regularized losses. Compared to a pure Mask R-CNN model, the overall algorithm can achieve equivalent performance in terms of accuracy and completeness. However, unlike Mask R-CNN that produces irregular footprints, our framework generates regularized and visually pleasing building boundaries which are beneficial in many applications.