Abstract

As video volumes grow, analysts are increasingly able to query the real world. Since manually watching these growing volumes of video is infeasible, analysts have increasingly turned to deep learning to perform automatic analyses. However, these methods are: costly (running up to 10x slower than real time, i.e., 3 fps) and cumbersome to deploy, requiring writing complex, imperative code with many low-level libraries (e.g., OpenCV, MXNet). There is an incredible opportunity to leverage techniques from the data management community to automate and optimize these analytics pipelines. In this paper, we describe our ongoing work in the Stanford DAWN lab on BlazeIt, an analytics engine for scalable and usable video analytics that currently contains an optimizing query engine. We propose a demonstration of BlazeIt’s query language, FrameQL, its use cases, and our preliminary work on debugging machine learning, which will show the feasibility of video analytics at scale. We further describe the challenges that arise from large-scale video, progress we have made in automating and optimizing video analytics pipelines, and our plans to extend BlazeIt.