Demuxed 2018
2018-10-17..2018-10-18
San Francisco, United States
Demuxed ·
[Demuxed 2015, Demuxed 2016, Demuxed 2017, Demuxed 2018, Demuxed 2019, Demuxed 2020, Demuxed 2021, Demuxed 2023, Demuxed 2024, Demuxed 2025]
| title | author | length |
|---|---|---|
| Lessons learned from streaming the World Cup in UHD with HDR | Billy Romero, Thomas Symborski | 23m |
Launches a live UHD HDR10 workflow in 36 days using HEVC input from Fox (270 Mbps transport), encoder/packager Titan Live on AWS c5.18xlarge, fragmented MP4 /… | ||
| Narrowing the Native Gap | James Yeh | 10m |
Compares HTML5, native (AV player on iOS, exoplayer on Android) and a hybrid using native playback with an HTML5 UI, with measured metrics rather than… | ||
| What to do after per-title encoding | Ben Dodson, Nick Chadwick | 21m |
Per-title becomes a pipeline: build convex hulls with VMAF, target VMAF 94 for the top rendition, and train classifiers plus a convolutional LSTM to predict… | ||
| Subjective video quality assessment for mobile devices | Sebastiaan Van Leuven | 10m |
Claim: Adaptive Paired Comparison produces fast, dependable mobile video quality scores by showing each test clip once and using Adaptive Paired Comparison… | ||
| Introducing dav1d, a new AV1 decoder | Jean Baptiste, Ronald S. Bultje | 12m |
dav1d is an open source AV1 decoder under BSD license that targets being small, fast, and cross‑platform. It is written in C plus hand‑written assembly… | ||
| An FFmpeg maintainer's deep dive: iOS VideoToolbox and Android MediaCodec integration | Aman Gupta | 11m |
FFmpeg exposes iOS VideoToolbox and Android MediaCodec so hardware H.264 decoding returns platform-specific frames instead of raw YUV buffers, enabling a… | ||
| VVC - the Next-Next Gen Codec | Christian Feldmann | 24m |
Progress on Versatile Video Coding is concrete: the VTM v2 reference encoder VTM v2 shows ~23% bitrate reduction versus HEVC on random access tests, with… | ||
| Media innovation in minutes | Naveen Mareddy | 19m |
A MapReduce style platform called Archer runs user-supplied Docker images and ffmpeg recipes at serverless scale so hundreds of thousands of CPUs finish… | ||
| Parsing CEA-608 Captions from FMP4s | Lahiru Dayanada | 08m |
Explains how CEA-608 captions are actually extracted from FMP4s by locating SEI NAL unit type 6 inside mdat boxes, mapping those NAL byte offsets to track… | ||
| Multi-CDN Jumpstart | David Hassoun | 16m |
Claims a Multi‑CDN setup eliminates single points of failure and boosts global performance while noting measurable tradeoffs: choices between DNS… | ||
| CHUNKY MONKEY | Will Law | 36m |
Claims that chunked-encoded CMAF delivers practical low-latency OTT at scale and shows concrete numbers: playback can hit around 500 milliseconds at the edge… | ||
| Time Machine: Reconstructing perceptually, during playback, detail lost in encoding | Fabio Sonnati | 25m |
Claims real-time perceptual reconstruction of high-frequency detail lost to lossy codecs by running a stack of GPU filters and a lightweight ML model during… | ||
| A Decentralized Fabric for Video over the Internet | Michelle Munson | 33m |
A decentralized software fabric overlays the Internet to store media as hashed binary parts, metadata and code and to compose outputs just-in-time composition… | ||
| Video, evolution, and gravity: how science affects digital video | Jon Dahl | 26m |
Asserts that many video decisions trace to biology and physics: human field of view ≈200°×135° and a fovea the size of a thumb at arm’s length, John’s law of… | ||
| Identifying Root Cause via Playback Metrics | Matt Fisher | 34m |
Playback metrics identify root cause fast when backed by concrete signals and tools: an in-house playback metrics pipeline plus session-level tooling lets… | ||
| LHLS(.js): Why Hls.js is standardizing low-latency streaming | John Bartos | 12m |
Claim: LHLS reduces latency from ~30s to ~2s while keeping existing encoders, CDNs and MSE players. Implementation: Hls.js implements a chunk-transfer pipeline… | ||
| Weaver: Inserting Ads in a Live Stream | Luke Curley | 10m |
A production system called Weaver stitches ad and live HLS playlists server-side at Twitch scale, serving per-user playlists so each viewer gets targeted ads… | ||
| Simulating livestreams for testing | Michael Roca | 08m |
Brightcove ships an open source tool called Murphy that actually converts HLS VOD master manifests into live streams on the fly, parsing renditions and VTT… | ||
| Digital Rights Management: The Good, The Bad, and The Ugly | Thasso Griebel | 37m |
State of DRM today: concrete workflows, failure modes, and mitigations. Covers CDMs and standards (EME, Widevine PlayReady FairPlay), init data → license… | ||
| How to be data-driven when you aren’t Netflix (or even if you are) | Charles Sonigo | 27m |
Built an analytics pipeline that scales with a tiny team, using Kafka and Druid and Flink plus Hadoop, Zeppelin and PySpark to collect and explore metrics… | ||
| Improving ABR at Netflix | Te-Yuan | 03m |
Netflix processes millions of playback sessions across devices and networks to measure where ABR decisions fail, finds that computing the true optimal… | ||
| The Dark Truths of a Video Dev Cartoon | Charles Sonigo | 07m |
A claim surfaces that Big Buck Bunny drives massive test traffic — 200,000 engineers rewatch the first 30 seconds over 20 million times a year — and one study… | ||
| Sub-frame Latency | Kieran Kunhya | 04m |
A complete capture-to-playback stack encodes and decodes video 16–32 lines at a time so presentation happens at sub-frame granularity, giving latency measured… | ||
| Just realized we had a “Time Machine” after Demuxed’18 day 1 | Drake Guan | 03m |
Real claim: a perceptual streaming operation system stores and serves low perceptual-quality video inside HD files so ops cost drops without changing delivered… | ||
| Webamp: Winamp in your browser | Tracey Jaquith | 03m |
A browser reimplementation called Webamp runs Winamp 2.9 in HTML5 and JavaScript rather than emulation and is available from the author’s repository. It is… | ||
| Why build a new native player? | Lars-Erik Ravn | 04m |
A new native video pipeline is built from scratch and claims hardware-accelerated native pipeline with a unified API across platforms, supports HLS DASH MP4 on… | ||
| Video Engineer Toolbox | David Hassoun | 05m |
Quick checklist of what actually helps: packet inspection with Charles and Wireshark for MITM and pcap captures, a fieldable mocap Raspberry Pi proxy that logs… | ||
| Seamless client-side ad insertion with ExoPlayer | Andrew Lewis | 03m |
Two common C-SAI patterns cause problems: pausing the player to fetch an ad causes a no rebuffering spinner moment, and the dual-player background-load… | ||
| Testing your Fastly VCL | Matt Ward | 02m |
A small test harness runs Fastly VCL against real services or a sandbox and claims to let you iterate on configs end to end rather than just eyeballing them… | ||
| A status update on SRT | Marc Cymontkowski | 04m |
Secure Reliable Transport claims sub‑one‑second real‑time delivery with preserved source timing so VBR video arrives without reshaping, and the open source… | ||
| Demuxed 2018 Video & Streaming Architecture | Phil Cluff | 02m |
Conference streaming architecture centers on handling 1080i and 1080p feeds through an HDMI into a Decimator then SDI into an ATEM control switcher that drives… | ||
| Watch Bots | Matthew McClure | 02m |
Lightning Talk #11!
Presented at Demuxed 2018 in San Francisco. | ||
| Magic Mirror: Understanding the Diversified Network Conditions of Twitch Community | Ying Cheng | 26m |
Claims an unsupervised pipeline that turns client-side QoS time series into a global dictionary of playback behavior via hierarchical clustering and dynamic… | ||
| Synchronize your Watches: Cross-platform stream synchronization of HLS and DASH | Seth Madison | 22m |
Philo builds synchronized playback for HLS and DASH by reusing the existing manifest timeline and broadcasting a UTC reference time and position so clients… | ||
| Machine Learning for ABR in production | Steve Robertson | 23m |
YouTube integrates machine learning into ABR incrementally rather than replacing it, deploying a v1 in production serving v1 at ~1 Gbps and targeting… | ||
| Towards Measuring Perceptual Video Quality & Why | Vasavee Vijayaraghvan | 15m |
Video quality measurement is an essential step in any media pipeline. There are quite a few objective metrics available today including well known SSIM & PSNR… | ||
| Virtues of Recycling in Multi-rate Encoding | Alex Giladi | 11m |
Claim: double the x265 encoding speed while keeping the same encoder. Method: a hierarchical pipeline encodes low resolution or higher‑bitrate copies first… | ||
| How to build a high quality 'live' HD VP9 Encoder | Avinash Ramachandran | 29m |
Claims a working FPGA-based VP9 encoder that delivers live savings close to reported VOD gains (YouTube ~50% offline, Netflix ~36%) by cutting bitrate roughly… | ||
| An engineer's perspective on EDLs | Megha Manohara | 14m |
EDLs function as blueprints that map sources to timelines and Netflix wrestles with a proliferation of formats — CMX3600, FCP XML and AAF — plus tooling like… | ||
| Virtues of Recycling in Multi-rate Encoding | Alex Giladi | 11m |
Claims a practical method to double x265 encoding speed by reusing motion vectors and quadtree structure from lower-resolution or higher-bitrate encodes, then… | ||
| nothing matches | ||