Presented By O’Reilly and Cloudera
Make Data Work
September 11, 2018: Training & Tutorials
September 12–13, 2018: Keynotes & Sessions
New York, NY

BlazeIt: An exploratory video analytics engine

Daniel Kang (Stanford University)
11:20am–12:00pm Wednesday, 09/12/2018
Data science and machine learning
Location: 1A 08 Level: Advanced
Secondary topics:  Media, Marketing, Advertising
Average rating: ****.
(4.00, 2 ratings)

Who is this presentation for?

  • Engineers and engineering managers

Prerequisite knowledge

  • A basic understanding of deep learning, particularly in relation to computer vision

What you'll learn

  • Explore BlazeIt, a video analytics engine that offers fast execution of queries over video


As video volumes grow, it’s increasingly important to prioritize human attention. Thus, practitioners have turned to automatic methods of processing data. While delivering impressive levels of accuracy, these automatic methods are infeasible to run at scale (10x slower than real time on an NVIDIA P100 GPU). Even if these methods can scale, they are difficult to deploy, requiring ad hoc solutions.

Daniel Kang offers an overview of exploratory video analytics engine BlazeIt, which offers FrameQL, a declarative SQL-like language for querying video, and a query optimizer for executing these queries. You’ll see how FrameQL can capture a large set of real-world queries ranging from aggregation (e.g., counting cars) and scrubbing (e.g., finding interesting clips in video) and how BlazeIt can execute certain queries up to 2,000x faster than a naive approach.

Photo of Daniel Kang

Daniel Kang

Stanford University

Daniel Kang is a PhD student in the Stanford InfoLab, where he is supervised by Peter Bailis and Matei Zaharia. Daniel’s research interests lie broadly at the intersection of machine learning and systems. Currently, he is working on deep learning applied to video analysis.