Skip to content

Visual background manipulation

Dirk Holtwick edited this page Jun 18, 2020 · 1 revision

All manipulation happens locally!

Google's Javascript TensorFlow/BodyPix project is used to identify persons shapes. Based on this data the original video stream is used to write the manipulated image to a hidden canvas element. This is then used to create a new video stream, where the audio data is mixed in for transmission.

For background images the Unsplash.com service is used. A random image that matches the search term background will be proposed and is then downloaded and stored locally.

Clone this wiki locally