I. Introduction
Road maps play an important role in traffic management, vehicle navigation, urban planning, and other fields [1]. Road graph extraction from remote sensing images obtains the vectorized representation of road maps that regards the road as a graph structure composed of vertices and edges [2]. Annotating road graphs manually from the remote sensing images is usually time-consuming and labor-intensive. As a result, to obtain accurate and timely-updated road maps, automatic road graph extraction has become an emerging research topic in the remote sensing field that has attracted increasing attention in recent years.