Distilled Roads: Generalisable Road Network Extraction Across Sensors, Resolutions, and Region

2026-08-04

AI Summary

This work addresses the limited generalization of road extraction from satellite imagery caused by geographic variability, occlusions, and differences in sensor types and resolutions. To tackle this challenge, the authors propose a universal road extraction method that frames global road extraction as a continual adaptation task. By integrating curriculum learning with decreasing resolution, cross-resolution knowledge distillation, multi-sensor joint training, and a topology-aware loss function, the approach effectively mitigates catastrophic forgetting without relying on complex model architectures. Evaluated on both City-Scale and Global-Scale benchmarks, the method achieves up to a 22-point improvement in F1 score and a 15-point gain in APLS, while offering a threefold increase in inference speed, substantially outperforming existing state-of-the-art approaches.

Abstract

Road network segmentation from satellite imagery remains challenging due to large geographic variation in road appearance, occlusions, and domain shifts introduced by differing resolutions and sensors. Existing models, typically trained under narrow resolution--region combinations, generalise poorly to unseen environments such as rural settings, regions with distinct road materials, or imagery from new satellite platforms, often producing broken or disconnected predictions. Adapting these models to new domains usually requires retraining or fine-tuning, which is costly and risks catastrophic forgetting. In this work, we reframe global road extraction as a continual adaptation problem rather than an architectural one. Our framework combines cross-resolution knowledge distillation across a resolution-decreasing curriculum, multi-sensor training, and topology-aware supervision, yielding a single model that generalises across $0.3-1.0$ m imagery from multiple satellite platforms across continents. On publicly available benchmarks, including City-Scale and Global-Scale, our model outperforms state-of-the-art results by up to $22$ F1 points and $15$ APLS points, while remaining the most efficient, with $3\times$ faster inference. Our results suggest that improved robustness across diverse sub-meter satellite imagery can be achieved through targeted training strategies, such as data curricula, distillation, and topology-aware losses, rather than increasingly complex architectures.

Problem

• road network extraction
• satellite imagery
• domain generalization
• resolution variation
• sensor diversity

Innovation

• knowledge distillation
• cross-resolution learning
• topology-aware supervision
• multi-sensor training
• continual adaptation