SF Video Tech: Watching Twitch Live Stream Scores with TensorFlow

Earlier this year, Demuxed held their SF Video Tech event in our SoMa Clubhouse. The event featured an exciting talk and near-perfect live demo from Mux’s Nick Chadwick. If you’d like to attend the next event in person, RSVP here.

At the end of last year, Mixer, a Twitch competitor, introduced a new feature: Hypezone. Hypezone is a channel that always broadcasts the Player Unknown’s Battlegrounds stream closest to winning, and has consistently been one of the most watched channels on Mixer. Inspired by the technical challenge, Nick built a simple version of Hypezone (Hypebot3000) that does the same thing for Twitch streams.

In this talk, Nick outlines how he approached solving this problem, from efficiently watching twitch streams and struggling with OCR, to finally creating a working version using TensorFlow. He goes over the basics of screen capturing live streams, pre-processing them, and implementing a simple TensorFlow model to extract the current game score. And he does it all live!

Interested in joining Heavybit? Our program is the only one of its kind to focus solely on taking developer products to market. Need help with developer traction, product market fit, and customer development? Apply today and start learning from world-class experts.