A Quality-of-Experience Database for Adaptive Omnidirectional Video Streaming

Abstract

Recent advances in virtual reality (VR) technologies and devices have enabled new forms of media content, such as omnidirectional video (ODV) that attracts increasing attention of both academic and industrial communities. Omnidirectional video, which is also called $360^\circ$ video, represents panoramic spherical video that can give users an immersive viewing experience. Compared with traditional 2D video, the complex characteristic (high resolution, bandwidth intensive, etc.) of ODVs brings up new challenges to stream them under volatile network conditions and model the quality-of-experience (QoE) of end-users. To meet the requirements of practical VR applications, it is desired to design effective methods for adaptive bitrate streaming (ABR) and QoE evaluation of ODVs. In this article, we establish a large omnidirectional video streaming QoE database (VRQoE-JUFE), containing 1,440 adaptive streaming ODVs generated with diverse content. A comprehensive subjective experiment is conducted, where viewing behaviors and human opinions of total 180 subjects are collected. We provide a thorough statistical data analysis and carry out performance evaluation of the-state-of-art objective QoE models on the proposed database. The results suggest that QoE modeling for ODV streaming is an extremely challenging problem and there is a large space for improvement. Many interesting observations are made that may shed light on the improvement in both omnidirectional video QoE modeling and ABR strategies in the future. The annotated dataset from the tests is made publicly available for the research community.