PolyWorld: Polygonal Building Extraction with Graph Neural Networks in Satellite Images

While most state-of-the-art instance segmentation methods produce binary segmentation masks, geographic and cartographic applications typically require precise vector polygons of extracted objects instead of rasterized output. This paper introduces PolyWorld, a neural network that directly extracts...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Proceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) S. 1938 - 1947
Hauptverfasser: Zorzi, Stefano, Bazrafkan, Shabab, Habenschuss, Stefan, Fraundorfer, Friedrich
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.06.2022
Schlagworte:
Buildings
Codes
Computer network reliability
Computer vision
grouping and shape analysis
Image segmentation
Photogrammetry and remote sensing; Deep learning architectures and techniques; Segmentation
Predictive models
Satellites
ISSN:1063-6919
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:While most state-of-the-art instance segmentation methods produce binary segmentation masks, geographic and cartographic applications typically require precise vector polygons of extracted objects instead of rasterized output. This paper introduces PolyWorld, a neural network that directly extracts building vertices from an image and connects them correctly to create precise polygons. The model predicts the connection strength between each pair of vertices using a graph neural network and estimates the assignments by solving a differentiable optimal transport problem. Moreover, the vertex positions are optimized by minimizing a combined segmentation and polygonal angle difference loss. PolyWorld significantly outperforms the state of the art in building polygonization and achieves not only notable quantitative results, but also produces visually pleasing building polygons. Code and trained weights are publicly available at https://thub.com/zorzis/yWorl-PoldPretrainedNetwork.
ISSN:1063-6919
DOI:10.1109/CVPR52688.2022.00189