MadEye - Boosting Live Video Analytics Accuracy with Adaptive Camera Configurations

Explore a groundbreaking approach to enhancing live video analytics accuracy through adaptive camera configurations in this conference talk from NSDI '24. Delve into the innovative MadEye system, developed by researchers from Princeton University and Rice University, which automatically adjusts camera orientations to maximize accuracy while optimizing resource usage. Learn how MadEye's novel search algorithm and knowledge distillation strategy work together to efficiently explore and select the most effective camera orientations for diverse workloads. Discover the significant improvements achieved by this system, including accuracy boosts of up to 25.7% and resource cost reductions of 2-3.7 times. Gain insights into the potential impact of this technology on various applications of live video analytics.

NSDI '24 - MadEye: Boosting Live Video Analytics Accuracy with Adaptive Camera Configurations