SENSEI: Aligning Video Streaming Quality with Dynamic User Sensitivity
summary
Shows that per-video crowdsourced sensitivity profiles let adaptation trade short-term rebuffering for higher quality at important moments, implemented in a system called SENSEI that augments existing ABR algorithms. Uses two-step crowdsourcing (grouping similar chunks then sampling representatives) plus pruning and quality controls, yields $31.4/minute cost, boosts QoE prediction accuracy up to 37.1% and delivers 15.1% QoE improvement or the same QoE with ~26.8% less bandwidth.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:36Overview
- 1:02Conventional wisdom
- 2:44Which video has better quality
- 2:57User sensitivity
- 4:07Three questions
- 4:29Experiment
- 5:34Integration
- 6:04Video sentencing models
- 7:01Questioning cloud workers
- 7:52Twostep scheduling
- 9:24SENSEI vs traditional ABR
- 10:20Experimental setup
- 10:41Experimental results
- 11:27Crowdsourcing costs
- 11:40Summary
related talks
22mProviding better video experiences for the next billion users.
10mGoing Beyond Bitrate: Adaptation Based on Video Context
21mLearning in situ: a randomized experiment in video streaming
15mTowards Measuring Perceptual Video Quality & Why
14mAnalyzing Video Metrics like Richard Feynman
17mYuZu: Neural-Enhanced Volumetric Video Streaming
30mMLDS 2019: Anubhav Shrivastava of Voot at Machine Learning Developers Summit 2019
14mUsing Performance Signals to Improve Video Recommendations