A Full- and No-Reference Metrics Accuracy Analysis for Volumetric Media Streaming

Abstract

Volumetric media streaming will be one of the fundamental technologies to enable near future immersive multimedia experiences. In it, objects represented as sets of points (i.e. point-clouds), are presented to remote users wearing Head-Mounted Displays (HMDs). Due to the stringent bandwidth and latency requirements of such applications, small changes in the network can affect the user in unexpected manners (physical discomfort, lack of concentration, etc.). Therefore, there is a need for assessing the perceived quality of this type of applications in real-time, i.e, the Quality of Experience (QoE). Given that subjective evaluations are not feasible for (near) real-time applications, objectively measuring this quality will be a must. While traditional objective metrics could potentially be used to fulfill the task, it is still unclear how accurate they are to assess volumetric media. To this end, this paper presents a thorough correlation analysis of both Full Reference (FR) and No Reference (NR) objective metrics to subjective Mean Opinion Scores (MOS) for different volumetric streaming scenarios. To enhance the accuracy, multiple Region-Of-Interest (ROI) selection and weighting procedures have been applied and their influence on the results have been investigated. Our results show that the classical video quality metric Video Multimethod Assessment Fusion (VMAF) is well-suited as an objective benchmark for volumetric media streaming in terms of correlation to subjective scores, while a combination of NR features could provide a suitable real-time assessment. Finally, ROI selection proves to widen the range of objective metrics, which is an important issue to apply traditional objective metrics to volumetric media.