Video is one of the fastest-growing sources of data with rich semantic information and advances in deep learning have made it possible query this information with near-human accuracy. However, inference remains prohibitively expensive: the most powerful GPU cannot run the state-of-the-art at real time. In response, we present NoScope, which run queries over video 1000x faster.
NoScope achieves such speeds by exploiting redundancies in video: temporal, environmental, and query-specific redundancies. I will explain how we exploit these redundancies and explain how these concepts can be generalized.
NoScope is a new project in the Stanford InfoLab under Professors Matei Zaharia and Peter Bailis. John Emmons and Firas Abuzaid also contributed to the work. The code is open-sourced: https://github.com/stanford-futuredata/noscope
Daniel is a PhD student in the Stanford InfoLab supervised by Peter Bailis and Matei Zaharia. His research interests lie broadly in the intersection of machine learning and systems. Currently, he is working on deep learning applied to video analysis.
Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?
Join the conversation here (requires login)
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • email@example.com