Perceptual Quality Assessment of Virtual Reality Videos in the Wild

TL;DR

The VR Video Quality in the Wild (VRVQW) database is constructed, containing 502 user-generated videos with diverse content and distortion characteristics, and an objective quality assessment model for VR videos based on pseudocylindrical representation and convolution is developed.

Abstract

Investigating how people perceive virtual reality (VR) videos in the wild (i.e., those captured by everyday users) is a crucial and challenging task in VR-related applications due to complex authentic distortions localized in space and time. Existing panoramic video databases only consider synthetic distortions, assume fixed viewing conditions, and are limited in size. To overcome these shortcomings, we construct the VR Video Quality in the Wild (VRVQW) database, containing $502$ user-generated videos with diverse content and distortion characteristics. Based on VRVQW, we conduct a formal psychophysical experiment to record the scanpaths and perceived quality scores from $139$ participants under two different viewing conditions. We provide a thorough statistical analysis of the recorded data, observing significant impact of viewing conditions on both human scanpaths and perceived quality. Moreover, we develop an objective quality assessment model for VR videos based on pseudocylindrical representation and convolution. Results on the proposed VRVQW show that our method is superior to existing video quality assessment models. We have made the database and code available at https://github.com/limuhit/VR-Video-Quality-in-the-Wild.

Figures (10)

• Figure 1: Illustration of how people explore VR videos in the proposed VRVQW database. Under varying viewing conditions (e.g., starting points and exploration times), users may exhibit different viewing behaviors in the form of scanpaths, leading to different portions of the video being explored. As user-generated VR videos often come with localized authentic distortions, the perceived quality may vary with user viewing behaviors constrained by viewing conditions. Therefore, the incorporation of viewing conditions would be the key to the success of computational quality prediction of user-generated VR videos.
• Figure 2: 360° video processing pipeline, from optical acquisition to content consumption via an HMD. The optical acquisition and stitching are two main steps for 360° video creation, where authentic distortions arise.
• Figure 3: Visual examples of authentic distortions in VRVQW.
• Figure 4: We consider two types of starting points. Starting Point i@ (denoted by the light orange dot) and Starting Point ii@ (denoted by the dark green dot) offer poor and good initial viewing experiences, respectively. The video name in VRVQW is "D_ConfucianTemple".
• Figure 5: We consider two exploration times, one spanning the entire duration (i.e., 15 seconds) and the other set to the half of the former (i.e., 7 seconds). The initial viewport is from Starting Point i@. The video name in VRVQW is "F_BridegOpening2".