RexCam

This repo contains Python implementations of ReXCam: Resource-Efficient, Cross-Camera Video Analytics at Scale.

Dependencies

• PyTorch (0.4.0)
• torchvision (0.2.1)
• OpenCV ()
• AWS Deeplens ()
• MXNet ()

Python2 is recommended for current version.

Install

1. cd to the folder where you want to download this repo.
2. run git clone https://github.com/skybigzhou/RexCam.git.

Prepare data

Create a directory to store reid datasets under this repo via

Controller

Run the controller on remote laptop with

PYTHONPATH=. python src/controller/messenger_run.py

And

Inside messenger_run.py rememeber to modify spatial temporal correlation matrix corr_matrix, start_times, and end_times.

Then in function trigger() message sending format is (ip_address, instruction, keyword, task, model, video, bAnalysis, query, start_time, end_time)

Worker

Model Management

run model management

bash script.sh run modelmanagement

or

PYTHONPATH=. python src/worker/modelManagement.py -mps {list of model_path}

arguments: -mps model_path loading to GPU

Data Management

run data management

bash script.sh run datamanagement

or

PYTHONPATH=. python src/worker/dataManagement.py -t {duration_time}

arguments: -t duration time

Worker Thread

run worker for local version

PYTHONPATH=. python src/worker/script.py -f intel_mo_IR -mp /opt/awscam/artifacts/deploy_ssd_mobilenet_512 -t ssd -s AWSCAM -n deploy_ssd_mobilenet_512

arguments: -f {format} -mp {model_path} -t {task} -s {video source} -n {model_nickname}

run worker for remote control version

PTYHONPATH=. python src/worker/script.py