A prototype implementation of EQUI-VOCAL, which is a system to automatically synthesize compositional queries over videos from limited user interactions. See the technical report and the website for more details.

First, set up the backend code.

1. Download training videos at Dataset → Training Videos.

2. Create a new folder equi_app/static/equi_app/clevrer/, and place downloaded videos in it.

3. Ensure that the inputs/demo_queries_scene_graph directory is present in your locally cloned EQUI-VOCAL repository.

1. Modify config.json.

2. Run the following commands to set up the Django application:

python manage.py migrate
python manage.py collectstatic

3. Run the Django server:

python manage.py runserver