Non-central Panorama Indoor Dataset

We present the first dataset of non-central panoramas for scene understanding. Our dataset is composed of 2574 images of different Atlanta and Manhattan virtual environments. We also provide Depth maps for each image as well as different labels of the structural lines to obtain the layout of the environments.
This dataset can be used for scene understanding purposes, obtaining the layout and monocular depth estimation, as well as for evaluating geometrical solutions for non-central projection systems, since we provide 3D information of corners and lines of the environments and the images.