Road Extraction with Satellite Images and Partial Road Maps

Road extraction is a process of automatically generating road maps mainly from satellite images. Existing models all target to generate roads the scratch despite that large quantity maps, though incomplete, are publicly available (e.g. those OpenStreetMap) and can help with extraction. In this paper, we propose conduct based on images partial which new. We then two-branch Partial Complete Network (P2CNet) for task, has two prominent components: Gated Self-Attention Module (GSAM) Missing Part (MP) loss. GSAM leverages channel-wise self-attention module gate capture long-range semantics, filter out useless information, better fuse features branches. MP loss derived trying give more attention pixels do not exist in maps. Extensive experiments conducted demonstrate effectiveness our model, e.g. P2CNet achieves state-of-the-art performance IoU scores 70.71% 75.52%, respectively, SpaceNet OSM datasets.