Everest: Fast Deep Video Analytics

Object detection applications that are commonly used demand substantial time and GPU resources to process videos, which results in slow processing speeds and increased costs.

• Everest leverages sophisticated algorithms and probability theory to reduce the cost of using deep computer vision models.
• Everest is fully programmable and can be customized for a variety of tasks, eliminating the need for separate systems for different tasks.

• This technology enables users it to deliver results up to 10-20x faster (or equivalently, at 10-20x lower cost) than comparable systems that rely on GPUs.
• Everest has a wide range of use cases, including lost and found tracking for quickly locating misplaced items, sports analytics for real-time tactic reports during games.
• This technology can also be used in traffic planning for analyzing large volumes of traffic footage.

• SIGMOD Best Demo Paper Runner-Up
• Information of the award: https://cpii.hk/2022/06/05/acm-sigmod-2022/

• The system features a lost and found tracking function that enables quick location of misplaced belongings.
• The system can also be applied for sports analytics to generate real-time tactic reports during games.
• The system includes traffic planning function for analyzing large volumes of traffic footage
• With the one-click Programmable Controller, this system enables users to define 'Top-K' happy moments and quickly identify the most joyful parts in large volumes of traveling videos/vlogs.