Decoding the narrative in competitive Overwatch broadcasts

Code for the PyCon US 2020 presentation Decoding bias and narrative in competitive video games broadcasts with video analysis.

✨ Slides: https://aka.ms/pycon2020-overwatch

🗓 PyCon US schedule description: https://us.pycon.org/2020/schedule/presentation/107/

Installation 🏗

What you will need:

• VS Code + Python extension
• Python 3.X
• An Azure subcription with a Computer Vision resource (documentation)

> git clone git@github.com:kimadeline/overwatch-ocr.git
> cd overwatch-ocr
> python -m venv .venv
> source .venv/bin/activate
> python -m pip install -r requirements.txt

Manual work to be done (sorry) ✍️

• A list of video files for the games you want to parse, preferably without the breaks between maps. It's easier now that the OWL is on YT Gaming, otherwise here's the full USA vs China OWWC match, now go have fun trimming it;

• A JSON reference file mapping player names, team names and shorthand names, look at static/owwc_teams.json for an example.

• Create the following folders at the workspace root (siblings of src):

  • input: Where you'll put your video files
  • output: Where the algorithm will split the frames and crop them
  • data: Where we will store JSON files containing player names extracted from each frame, and also the reference teams file

Usage ⚒

Quickstart with existing OWL results data (PyCon 2020)

To visualize graphs for the OWL matches I used to compute the stats in my PyCon talk:

• Pull the owl_2020 branch of this repo
• Copy the owl_results/*_pov.json files to the data folder
• Update the arguments of the display_dashboard function in main.py
• ✨ python main.py ✨

Other OW broadcasts

• Copy your video files to the input folder
• Update the GAMES_LIST variable in src/__init__.py to be the list of the videos you want to parse (don't include the extension, but if it's not an mp4 file update L.12 of filename.py)
• Save your Azure subscription key and endpoint somewhere:
  • Option 1: Add your subscription key and your endpoint to a .env file at the root of the repo
  • Option 1: Write them directly in ocr.py
• Run everything (this might take a while because I make unoptimized, throttled Azure OCR calls)
  • If you followed option 1 from above, launch the Python: Main script configuration (you can't use the run button because of this issue)
  • Otherwise, use the run button in the top-right corner, or call python main.py directly
• ⚠️ Make sure your computer doesn't go to sleep
• Profit 🥳

You should end up with something like this: