360D: A dataset and baseline for dense depth estimation from 360 images

Authors
A. Karakottas
N. Zioulis
D. Zarpalas
P. Daras
Year
2018
Venue
1st Workshop on 360o Perception and Interaction, European Conference on Computer Vision (ECCV) , Munich, Germany, 8 – 14 September 2018
Download

Abstract

We present a baseline for 360o dense depth estimation from a single spherical panorama. We circumvent the unavailability of coupled 360o color and depth image datasets by rendering a high quality 360o dataset from existing 3D datasets. We then train a CNN designed speci_cally for 360o content in a supervised manner, in order to predict a 360o depth map from a single omnidirectional image in equirectangular format. Quantitative and qualitative results show the need for training directly in 360o instead of relying on traditional 2D CNNs.