Benchmark assesses complex polygon extraction from satellite images
PolyTopoBench: A Benchmark for Complex Vector Polygon Generation from Remote Sensing Imagery
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Turning satellite pictures into detailed shapes on maps is hard, especially for complicated shapes with holes or multiple parts. The authors created PolyTopoBench, a testing system to see how well current computer programs can draw these tricky shapes from satellite images. They checked 11 different methods on many types of land areas like buildings and roads. They found that while most methods do okay with simple shapes, they struggle a lot with complex ones, showing there's still room for improvement. This work helps developers know which methods fail and where to focus next.
What this means in practice
- •For mapping software engineers: Evaluate and compare tools that convert satellite images into accurate complex polygon maps, improving map data quality.
- •For urban planners: Use better polygon extraction methods to generate detailed land-use boundaries for city planning from aerial images.
Authors
Zeping Liu, Ni Lao, Weiwei Sun, Gil Wolff, Yiqun Xie, Liang Zhao, Junfeng Jiao, Gengchen Mai
Abstract
Vector polygon generation converts visual inputs, e.g., remote sensing (RS) images, into vectorized polygonal geometries, supporting applications such as autonomous driving, vector map construction, and remote sensing. Early pipelines predict raster masks and post-process them into polygons, which prevents end-to-end optimization and may miss small objects or introduce inaccurate vertices. Recent methods directly generate vector polygons, but most focus on simple exterior contours, while they either cannot represent complex polygons with holes or fail to preserve their topology. In this paper, we propose PolyTopoBench, a unified evaluation framework for vector polygon generation from RS images with explicit emphasis on complex polygons. PolyTopoBench evaluates both exterior and interior rings, and benchmarks 11 representative methods, including segmentation-based polygonization pipelines, vision foundation model baselines, and specialized vector polygon generators, on two RS-image datasets covering buildings, roads, vegetation, and unvegetated regions. Experiments show that existing methods often recover simple exterior boundaries but degrade substantially on polygons with holes or multiple rings. These results reveal complex polygon generation as an unresolved challenge and motivate topology-aware benchmarks and model designs. Code and data are available at https://github.com/seai-lab/PolyTopoBench.