TensorFlow.js in love with YouTube and Chrome extension

It’s a ‘just for fun’ project for motion tracking in YouTube videos.
Available in the Chrome Web Store.

Link: YouTube™ motion tracking - Chrome Web Store

Some example animations: Watch motion tracking - YouTube
I am grateful for any feedback.

And so far I have only tested it with MacPro 2021. Everything runs very smoothly. Maybe someone has older hardware available for testing and can write something about the performance.


Due to other ML topics and time constraints I have to stop working on this experiment.
It was interesting to see what is currently possible.
Real-time motion detection in the browser for YouTube videos without external API calls.:fire:
A more useful use case for this should still be designed.
Many thanks to the TFJS team for making this experiment possible.:+1:

A demo video of the latest version is available here.

If you want, try the extension from Chrome Web Store for free.

The code is available on GitHub.

If you want to create your own browser extension, it’s easy with TensorFlow.js.
There are only three steps necessary.

  • Import the libraries
  • Create a detector
  • Run inference
    See MoveNet