judo-footage-analysis

This repository is work supporting "Semi-Supervised Extraction and Analysis of Judo Combat Phases from Recorded Live-Streamed Tournament Footage". The goal of the project is to automate live-stream recording segmentation into individual matches, extract combat phases from matches, and to gather statistics at the tournament level.

This project was done as part of CS8813 Introduction to Research at Georgia Tech Europe during the Spring 2024 semester.

quickstart

Checkout the repo and install any dependencies you may need to a virtual environment:

git checkout ${repo}
cd ${repo_name}

python -m venv .venv
pip install -r requirements.txt
pip install -e .

Install any of the relevant tools for running workflows:

• ffmpeg
• b2-tools
• google-cloud-sdk

running a workflow

Most of the data processing workflows are written as luigi scripts under the judo_footage_analysis/workflow module. These can be run as follows:

# in a terminal session
luigid

# in a separate session
python -m judo_footage_analysis.workflow.{module_name}

You can watch the progress of a job in the terminal or from the luigi web-ui at http://localhost:8082.