Current Labeling PracticesĬurrent approaches to road labeling are often manually intensive. In the following post we describe our new evaluation metric in greater detail. As we will show below, the commonly used pixel-based F1 score is suboptimal for our task this leads us to develop a new metric motivated by graph theory concepts that we believe will focus competitors on routing rather than just static road pixel identification. In order to make significant advances in automated inference of routing information from satellite imagery, we recognize that metric selection for our competition is critical. Yet while satellites in theory provide an optimal method to rapidly obtain relevant updates, most existing computer vision research methods for extracting information from satellite imagery are neither fully automated, nor able to extract routing information from imaging pixels.Īccordingly, CosmiQ Works, in collaboration with SpaceNet partners DigitalGlobe, NVIDIA and hosted on AWS, plan to launch a third SpaceNet challenge focused on determining road networks and routing information directly from satellite imagery.
Spacenet vegas update#
Looking ahead, we believe the frequent revisits of satellite imaging constellations may accelerate existing efforts to quickly update road network and routing information.
Spacenet vegas manual#
Existing data collection methods such as manual road labeling or aggregation of mobile GPS tracks are currently insufficient to properly capture either underserved regions with infrequent data collection, or the dynamic changes inherent to road networks in rapidly changing environments. Autonomous vehicles is one of many examples on the commercial front, as self-driving cars rely heavily upon highly accurate road maps. In the humanitarian realm, for example, traveling to disaster stricken areas can be problematic for relief organizations, particularly if flooding has destroyed bridges or submerged thoroughfares.
SpaceNet Road Detection and Routing Challenge - Part Iĭetermining optimal routing paths in near real-time is at the heart of many humanitarian, civil, military, and commercial challenges.
0 kommentar(er)
