Presented By
O’Reilly + Cloudera
Make Data Work
March 25-28, 2019
San Francisco, CA
Please log in

Scanner: Efficient video analysis at scale

Fait Poms (Stanford University), Will Crichton (Stanford University)
11:50am–12:30pm Thursday, March 28, 2019
Data Engineering & Architecture
 Location: 2008
Secondary topics:  Deep Learning, Media, Marketing, Advertising
Average rating: ****.
(4.75, 4 ratings)

Who is this presentation for?

  • Data scientists and analysts, researchers, and distributed systems engineers

Level

Beginner

Prerequisite knowledge

  • Basic knowledge of Python and computer vision

What you'll learn

  • Explore Scanner, a new system for analyzing video data, as well as a wide range of video analysis and processing applications and their workload characteristics

Description

A decade ago, systems like Apache Spark and Hadoop made it possible to process big numerical and textual data workloads on clusters of hundreds of machines. Today, the majority of data in the world is video, and companies like Google and Netflix are starting to develop bespoke in-house solutions for processing this big video data at scale.

Alex Poms and Will Crichton offer an overview of Scanner, the first open source distributed system for building large-scale video processing applications. Scanner is being used at Stanford for analyzing TBs of film with deep learning on GCP and at Facebook for synthesizing VR video on AWS. Alex and Will begin by outlining the space of large-scale video analytics applications, and then dive deep into the specific workload characteristics of these applications and how they differ from traditional big data analytics. You’ll learn how Scanner is designed to address these workloads and discover detailed examples of existing applications written in Scanner that have been scaled out to hundreds of machines in the cloud.

Photo of Fait Poms

Fait Poms

Stanford University

Fait Poms is a CS PhD student at Stanford, where she is advised by Kayvon Fatahalian, as well as a research contractor for Oculus/Facebook. Fait’s PhD research focuses on designing algorithms and programmable systems for efficiently analyzing video. She has published and presented work at SIGGRAPH and CVPR on systems for large-scale video analysis and efficient 3D reconstruction using deep learning.

Photo of Will Crichton

Will Crichton

Stanford University

Will Crichton is a PhD student in computer science at Stanford, advised by Pat Hanrahan. He creates systems that merge research in parallel computing and programming language design to solve impactful problems. Will’s current focus is on tools to enable large-scale visual data analysis, or processing massive collections of images and videos, including published work at SIGGRAPH.