Efficient Measurement of Quality at Scale in Facebook Video Ecosystem

SPIE Optics + Photonics


This paper describes FB-MOS, a metric that is used to measure video quality at scale in FB Video Ecosystem. Facebook processes a very large number of videos daily that collectively receive billions of views each day and hence both the accuracy and computational complexity of the metric are equally important. As the quality of uploaded user-generated content (UGC) source itself varies widely, FB-MOS consists of both a no-reference metric component to assess input (upload) quality and a full-reference component to assess quality preserved in the transcoding and delivery pipeline. FB videos can be watched on a variety of devices (Mobile/Laptop/TV) in varying network conditions, and often switched between in-line view and full-screen view during the same viewing session. We show how FB-MOS metric can accurately account for all this variation in viewing condition while minimizing the computation overhead to offer such measurement. We also discuss how this metric allows for end-to-end quality monitoring at scale, as well as guide encoding and delivery optimizations. The paper also discusses some of the optimizations to enable its use to achieve real-time quality measurement for Live videos. Another aspect of the Facebook video product is the wide variation in popularity of videos where less popular UGC content may receive relative very few views while highly watched professional or viral UGC content can receive millions of views. We discuss how the computational overhead of the metric can scale with the popularity of video where more compute is expended on more popular videos to get more accurate metrics while spending less compute on less popular videos.

Related Publications

All Publications

HPCA - March 3, 2021

Heterogeneous Dataflow Accelerators for Multi-DNN Workloads

Hyoukjun Kwon, Liangzhen La, Michael Pellauer, Tushar Krishna, Yu-Hsin Chen, Vikas Chandra

MLSys - April 8, 2021

CPR: Understanding and Improving Failure Tolerant Training for Deep Learning Recommendation with Partial Recovery

Kiwan Maeng, Shivam Bharuka, Isabel Gao, Mark C. Jeffrey, Vikram Saraph, Bor-Yiing Su, Caroline Trippel, Jiyan Yang, Mike Rabbat, Brandon Lucia, Carole-Jean Wu

TSE - January 1, 2020

Approximate Oracles and Synergy in Software Energy Search Spaces

Bobby R. Bruce, Justyna Petke, Mark Harman, Earl T. Barr

OOPSLA - October 25, 2019

Getafix: Learning to Fix Bugs Automatically

Johannes Bader, Andrew Scott, Michael Pradel, Satish Chandra

To help personalize content, tailor and measure ads, and provide a safer experience, we use cookies. By clicking or navigating the site, you agree to allow our collection of information on and off Facebook through cookies. Learn more, including about available controls: Cookies Policy