VLocNet++: Deep Multitask Learning for Semantic Visual Localization and Odometry | IEEE Journals & Magazine | IEEE Xplore