Delivering better video encodes for legacy devices
summary
Netflix replaces PSNR with VMAF for per-title complexity decisions, adds per-chunk bitrate variations via multi-pass encodes and retunes bitrate ladders, producing ~BD-rate 6.5% median gains (top 10% ~16%). A/B experiments show average session bitrate down over 20%, rebuffer per hour ~10% lower and CDN footprint ~20% smaller. Changes are productized, rolled into re-encodes, and A/B testing surfaces any device-specific decoding failures for follow-up.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:23About Netflix
- 1:19Encoding Pipeline
- 2:59Nonupgradable devices
- 4:12Goals
- 5:11Changes
- 5:59Productization
- 6:43VMAF
- 8:14Complexitybased encoding
- 9:06Example
- 10:39Aggregate rate distortion performance
- 11:51Overall bitrate ladder
- 12:52AB testing
- 14:11Estimated benefits
- 14:51Delivery infrastructure
- 16:06Conclusion
related talks
10mGoing Beyond Bitrate: Adaptation Based on Video Context
20mNotes on AV1 productization and royalty-free video codecs outlook
20mStreaming at 250 kbps: raise your expectations
27mHighly-Efficient SVT-AV1-based Solutions for VOD Applications
11mFrom Artifacts to Art: A Data-Driven Guide to Android Video Encoding
03mImproving ABR at Netflix
31mAV1 Update
45mNetflix's success: Combining collaboration, hardware monitoring & AI