SAM-Road is an adaptation of the Segment Anything Model (SAM) for extracting large-scale, vectorized road network graphs from satellite imagery. The model leverages SAM's semantic segmentation capabilities to predict graph geometry by generating dense segmentation masks for roads and intersections, then extracting vertices via non-maximum suppression. For graph topology, a lightweight transformer-based graph neural network is used to estimate edge existence probabilities between vertices. SAM-Road directly predicts graph vertices and edges without complex post-processing, achieving high accuracy and speed. It outperforms state-of-the-art methods like RNGDet++ in speed, being 40 times faster on the City-scale dataset. The model's simple design allows it to produce accurate road network graphs for large regions, with applications in navigation, autonomous vehicles, and urban planning. SAM-Road uses a pre-trained SAM image encoder, a geometry decoder for mask prediction, and a topology decoder for edge prediction. It employs a sliding-window approach for large regions, enabling efficient inference and parallel processing. The model is evaluated on the City-scale and SpaceNet datasets, demonstrating high accuracy in both geometry and topology prediction. SAM-Road's performance is attributed to the power of the foundational vision model SAM, which enables precise geometry prediction and effective graph representation. The model's efficiency and accuracy make it a promising solution for road network graph extraction from satellite imagery.SAM-Road is an adaptation of the Segment Anything Model (SAM) for extracting large-scale, vectorized road network graphs from satellite imagery. The model leverages SAM's semantic segmentation capabilities to predict graph geometry by generating dense segmentation masks for roads and intersections, then extracting vertices via non-maximum suppression. For graph topology, a lightweight transformer-based graph neural network is used to estimate edge existence probabilities between vertices. SAM-Road directly predicts graph vertices and edges without complex post-processing, achieving high accuracy and speed. It outperforms state-of-the-art methods like RNGDet++ in speed, being 40 times faster on the City-scale dataset. The model's simple design allows it to produce accurate road network graphs for large regions, with applications in navigation, autonomous vehicles, and urban planning. SAM-Road uses a pre-trained SAM image encoder, a geometry decoder for mask prediction, and a topology decoder for edge prediction. It employs a sliding-window approach for large regions, enabling efficient inference and parallel processing. The model is evaluated on the City-scale and SpaceNet datasets, demonstrating high accuracy in both geometry and topology prediction. SAM-Road's performance is attributed to the power of the foundational vision model SAM, which enables precise geometry prediction and effective graph representation. The model's efficiency and accuracy make it a promising solution for road network graph extraction from satellite imagery.