Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Per-Title Optimization 2.0

254 views

Published on

Learn what it takes to create target viewer quality based bitrate ladders for Live content and Video on Demand assets. Send the bits only when they are needed. Control viewer experience and efficiently stream a LIVE or VOD title with a ladder tailor-made to its specific characteristics and complexity.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Per-Title Optimization 2.0

  1. 1. Viewer Intelligence™ to Deliver The Ultimate Viewing Experience™ @SSIMWAVE | SSIMWAVE.com Per-Title Optimization 2.0 Zhou Wang PhD, FIEEE, FCAE Chief Science Officer
  2. 2. PER-TITLE OPTIMIZATION Adapting encoding bitrate for each title based on video content each channel Per-channel Optimization each scene Per-scene Optimization each GoP Per-GoP Optimization each frame Per-frame Optimization each block Per-block Optimization …… Content-Adaptive Encoding
  3. 3. PER-TITLE OPTIMIZATION Fixed-Bitrate Encoding Content-Adaptive Encoding reduced bitrate (on average) reduced quality variation  Intuition  Benefits “easier” content – use lower bitrate “more difficult” content – use higher bitrate
  4. 4. PER-TITLE OPTIMIZATION – FROM 1.0 TO 2.0 - Use intuitive/heuristic complexity measures - Shoot in the dark (make multiple bitrate attempts) - Verify results afterwards (subjectively/objectively) - Ad-hoc approach for cross-resolution assessment - Ad-hoc approach for cross-device assessment - Define clear optimization goal (cost function) first - Hit right on target (use quality-first principle) - Use comprehensive quality-rate-resolution model - Use consistent cross-resolution metric - Use consistent cross-device metric interactive, inaccurate, expensive automatic, accurate, low-cost Per-title Optimization 1.0 Per-title Optimization 2.0
  5. 5. PER-TITLE OPTIMIZATION 2.0 – QUALITY METRIC IS THE KEY TO SUCCESS Must-Have Properties - High accuracy, high speed - Easy-to-understand, easy-to-use - Fine granularity: per-title/scene/GoP/frame/block/pixel quality assessment - Consistent scoring across content, resolution, dynamic range, and viewing device/condition - Versatile: full-reference, no-reference, degraded-reference - End-to-end: consistent scoring between server and client
  6. 6. SSIMPLUS – HIGH ACCURACY Well aligned with MOS Correlation > 0.97 Only metric achieving 0.9 average correlation with MOS
  7. 7. SSIMPLUS – HIGH SPEED
  8. 8. SSIMPLUS – EASY-TO-UNDERSTAND, EASY-TO-USE quality meter Large-scale deployment, running 24/7
  9. 9. SSIMPLUS – FINE GRANULARITY test video frame pixel-precision quality map
  10. 10. SSIMPLUS – CONSISTENT SCORING cross user device cross dynamic range cross video content cross resolution/ frame rate
  11. 11. SSIMPLUS – HIGH VERSATILITY For source - Single-ended no-reference (NR) quality assessment For test - Double-ended full-reference (FR) perceptual fidelity assessment - Degraded-reference (DR) absolute quality assessment
  12. 12. SSIMPLUS – END-TO-END CONSISTENT QUALITY MONITORING MEASURE END-TO-END
  13. 13. CROSS-RESOLUTION QUALITY ASSESSMENT The ideal convex hull structure for rate-quality envelop Source: https://medium.com/netflix-techblog/per-title-encode-optimization-7e99442b62a2
  14. 14. 30 32 34 36 38 40 42 44 46 0 500 1000 1500 2000 2500 3000 3500 4000 234p 360p 432p 720p 1080p PSNR fails to produce meaningful convex hull structure CROSS-RESOLUTION QUALITY ASSESSMENT: PSNR
  15. 15. 10 20 30 40 50 60 70 80 90 100 0 500 1000 1500 2000 2500 3000 3500 4000 234p 360p 432p 720p 1080p VMAF fails to produce convex hull structure CROSS-RESOLUTION QUALITY ASSESSMENT: VMAF
  16. 16. 20 30 40 50 60 70 80 90 0 500 1000 1500 2000 2500 3000 3500 4000 360p 432p 576p 720p 1080p SSIMPLUS creates ideal convex hull structure CROSS-RESOLUTION QUALITY ASSESSMENT: SSIMPLUS
  17. 17. CONTENT VARIATION animation movie sports more difficulteasier Content not shown due to copyright
  18. 18. PER-TITLE ENCODING LADDER Name Frame Rate Aspect Ratio Resolution Target Quality Device Original Bitrate (kbps) ANIMATION MOVIE SPORTS New Bitrate Savings New Bitrate Savings New Bitrate Savings Profile 1 24 16x9 416x234 75 Phone 400 146 64% 310 23% 360 10% Profile 2 24 16x9 416x234 78 Phone 600 200 67% 458 24% 537 11% Profile 3 24 16x9 640x360 82 Phone 800 255 68% 607 24.1% 714 10.75% Profile 4 24 16x9 768x432 85 Phone 1400 383 73% 784 44% 1492 -6.57% Profile 5 24 16x9 1280x720 82 TV 2200 1033 53% 1846 16.1% 2279 -3.59% Profile 6 24 16x9 1280x720 85 TV 3400 1875 45% 2771 18.5% 3285 3.38% Profile 7 24 16x9 1920x1080 85 TV 3400 1201 65% 2455 27.8% 4575 -34.56% Profile 8 24 16x9 1920x1080 88 TV 5500 1960 64% 4000 27.3% 7774 -41.34% Average Saving = 60.2% 25.2% -18.7%
  19. 19. CONCLUSION  Per-title optimization 2.0 – Clearly defined optimization goal – Comprehensive quality-rate-resolution model – The quality-first principle – Automatic, accurate and low-cost  Quality metric is the key – Accuracy, speed, granularity, versatility, consistent scoring …...  Large average bit rate savings

×