Most autonomous vehicle designs rely on accurate 3d maps. In this talk, we describe an end-to-end mapping system for crowdsourcing maps with semantically meaningful objects such as traffic signs (6 dof) and traffic lanes (3d splines). For a stretch of road in San Diego, with just few journeys, we achieve absolute accuracy of 0.4m at any traffic sign corner and relative accuracy of 0.1m. Our system leverages precise real-time positioning, DNN based real-time landmark detection, clustering and computer vision geometry techniques for the construction of 3d maps. We show videos for visualizing the real-time outputs of our system as well as a visualization of the 3d map.