# Fast building segmentation from satellite imagery and few local labels

arXiv: https://arxiv.org/abs/2206.05377 · HTML: https://mlsanity.com/paper/2206.05377/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Builds on 11 of its 28 references that are in the map.
- First posted on arXiv 10 Jun 2022.

## Builds on (most cited first, 11 of 11)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Kornia: an Open Source Differentiable Computer Vision Library for PyTorch](https://mlsanity.com/paper/1910.02190/) (2019)
- [SpaceNet: A Remote Sensing Dataset and Challenge Series](https://mlsanity.com/paper/1807.01232/) (2018)
- [Image to Image Translation for Domain Adaptation](https://mlsanity.com/paper/1712.00479/) (2017)
- [xBD: A Dataset for Assessing Building Damage from Satellite Imagery](https://mlsanity.com/paper/1911.09296/) (2019)
- [Continental-Scale Building Detection from High Resolution Satellite Imagery](https://mlsanity.com/paper/2107.12283/) (2021)
- [ColorMapGAN: Unsupervised Domain Adaptation for Semantic Segmentation Using Color Mapping Generative Adversarial Networks](https://mlsanity.com/paper/1907.12859/) (2019)
- [TernausNetV2: Fully Convolutional Network for Instance Segmentation](https://mlsanity.com/paper/1806.00844/) (2018)
- [Assessing out-of-domain generalization for robust building damage detection](https://mlsanity.com/paper/2011.10328/) (2020)
- [A Semantic Segmentation Network for Urban-Scale Building Footprint Extraction Using RGB Satellite Imagery](https://mlsanity.com/paper/2104.01263/) (2021)
- [Scale Aware Adaptation for Land-Cover Classification in Remote Sensing Imagery](https://mlsanity.com/paper/2012.04222/) (2020)

## Built on by (most cited first, 9 of 9)

- [Evaluation Challenges for Geospatial ML](https://mlsanity.com/paper/2303.18087/) (2023)
- [Mask Conditional Synthetic Satellite Imagery](https://mlsanity.com/paper/2302.04305/) (2023)
- [Cross Pseudo Supervision Framework for Sparsely Labelled Geospatial Images](https://mlsanity.com/paper/2408.02382/) (2024)
- [Classifying Bicycle Infrastructure Using On-Bike Street-Level Images](https://mlsanity.com/paper/2410.19194/) (2024)
- [Seeing the roads through the trees: A benchmark for modeling spatial dependencies with aerial imagery](https://mlsanity.com/paper/2401.06762/) (2024)
- [Improving satellite imagery segmentation using multiple Sentinel-2 revisits](https://mlsanity.com/paper/2409.17363/) (2024)
- [Latent Domain Modeling Improves Robustness to Geographic Shifts](https://mlsanity.com/paper/2503.02036/) (2025)
- [Comparison of Segmentation Methods in Remote Sensing for Land Use Land Cover](https://mlsanity.com/paper/2507.18099/) (2025)
- [HASTE: A Platform for Rapid Post-Disaster Building Damage Assessment](https://mlsanity.com/paper/2607.11838/) (2026)

## Abstract (arXiv)

Innovations in computer vision algorithms for satellite image analysis can enable us to explore global challenges such as urbanization and land use change at the planetary level. However, domain shift problems are a common occurrence when trying to replicate models that drive these analyses to new areas, particularly in the developing world. If a model is trained with imagery and labels from one location, then it usually will not generalize well to new locations where the content of the imagery and data distributions are different. In this work, we consider the setting in which we have a single large satellite imagery scene over which we want to solve an applied problem -- building footprint segmentation. Here, we do not necessarily need to worry about creating a model that generalizes past the borders of our scene but can instead train a local model. We show that surprisingly few labels are needed to solve the building segmentation problem with very high-resolution (0.5m/px) satellite imagery with this setting in mind. Our best model trained with just 527 sparse polygon annotations (an equivalent of 1500 x 1500 densely labeled pixels) has a recall of 0.87 over held out footprints and a R2 of 0.93 on the task of counting the number of buildings in 200 x 200-meter windows. We apply our models over high-resolution imagery in Amman, Jordan in a case study on urban change detection.