The poses of these datasets are both obtained with SfM reconstructions or differential GPS that provides localization accuracies inside 5555 cm. All these datasets solely comprise street-degree photos taken from a digital camera mounted on a vehicle or a handheld system. In distinction, our DAG dataset incorporates photos taken by a floor car and an airplane. Aerial-to-Street-Level registration was addressed by Shan et al. Just like us, they suggest a view-dependent feature matching course of. However, they assume to know the approximate position (inside 20202020 meter) of the road level picture, whereas we first run a place recognition model to acquire this coarse localization. Another distinction to their work is that their 3D reconstruction is created from the street-level image, which is just attainable when multiple road-level photos of the identical area are available. We present that we can create an accurate 3D mannequin from aerial images and render street-stage photos. This generalizes to scenes with only few street-stage images of the scene.