Subjective Quality Assessment of User-Generated $360^{\circ}$ Videos

In this poster, we establish one of the largest virtual reality (VR) video database, containing 502 user-generated videos with rich content and commingled authentic distortions (often localized in space and time). We capture viewing behaviors (i.e., scanpaths) of 139 users, and collect their opinion scores of perceived qual-ity under four different viewing conditions (two starting points $\times$ two exploration times). We provide a thorough statistical analysis of recorded data, resulting in several interesting observations, which reveal how viewing conditions affect human behav-iors and perceived quality. The database is available at https://github.com/Yao-Yiru/VR-Video-Database.

1 Introduction

As virtual reality (VR) acquisition and display systems become widely accessible, people are getting used to capturing, editing, and interacting with VR content, which is evidenced by the accelerated proliferation of videos uploaded to popular video sharing and social media platforms (e.g., Bilibili and Youtube). A practical issue arising from user-generated videos is that they come with complex visual artifacts (i.e., the so-called authentic distortions) from acquisition to processing stages [2]. Thus, understanding how people perceive video distortions in virtual environments is central to many VR-enabled video applications.