VisionTracing

VisionTracing is a web application for running keypoint detection, tracking, and subject re-identification on video files.

Deployment

Docker

(Requires Docker to be installed.)

Run locally or in the cloud with:

docker-compose up

Heroku

(Requires a Heroku account and the Heroku CLI to be installed.)

Pick a suffix to differentiate this deployment from the one at visiontracing.herokuapp.com:

$ heroku apps:create visiontracing-<suffix>

Make sure that the Heroku remote was created:

$ git remote
heroku
origin

Configure the Heroku app stack to use Docker:

heroku stack:set container

Provision Redis:

$ heroku addons:create heroku-redis:hobby-dev

Provision Postgres:

$ heroku addons:create heroku-postgresql:hobby-dev

Provision at least one worker:

$ heroku ps:scale worker=1

Push to Heroku to deploy:

git push heroku master

Once deployment is complete, monitor the application logs:

heroku logs --tail

Usage

The following applies whether deployed on heroku (e.g. at http://visiontracing-[suffix].herokuapp.com) or docker-compose (e.g. at http://localhost:5000):

• Observe queues, workers, and jobs in the RQ-dashboard at the '/rq' endpoint.

• Tail logs to observe stdout from web and worker processes.

• Visit the top level '/' endpoint to enqueue a message. Refresh several times in rapid succession, then watch the RQ-dashboard and/or logs to confirm the queue is being drained by the workers.

TODO

This currently doesn't do anything useful. Next steps:

• Update web app to allow the user to upload a video file. See https://github.com/MLDSAI/ContactTracingAI.com/blob/master/app/views/video.py#L84 for an example of how to do this.
• Refactor code for detection, tracking, and subject re-identification in the visiontracing directory into something that can be run in a worker.
• Display the results to the user