/everest

Efficient Top-K Query in Video Analytics

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

Everest

This is the referece implementation of Everest described in:

[SIGMOD 2021] Top-K Deep Video Analytics: A Probabilistic Approach [arxiv]

Authors: Ziliang Lai, Chenxia Han, Chris Liu, Pengfei Zhang, Eric Lo, Ben Kao

Recommended Hardware

  • CPU: Intel i9-7900X or above
  • GPU: NVIDIA GTX1080Ti or above
  • RAM: >= 64GB

Setup

git clone https://github.com/everest-project/everest.git
cd everest
docker pull zllai/everest:1.2

Run a sample query

Query: find top-100 frames with largest number of cars in a 5-hour traffic footage

  1. Download the sample files:
./tools/download_sample.sh
  1. Run the container:
./tools/docker.sh
  1. Run the query:
python3 everest.py @config/traffic_footage.arg