🔮 Object-detection in the browser!
Using TensorFlow.js (see here for our blog posts on the topic) we can use the camera on your device to detect objects and draw boxes around them! In order to do this, we need to request to access your camera.
When it's working, you will see something like the below (depending on what it's pointing at!)
Where we have black boxes drawn around what it thinks it sees; with the name of the thing above it, and in brackets, (roughly) an estimate of "how confident" the network is that what it's it is seeing. Note that the network itself is not perfect, and it will get confused and certianly won't detect everything (for example, it misses the second cup in this photo, as well as the book). But it gives a feeling for the kinds of things that cna be done!
Demo
Loading ...
Detectable things
Details
This code is a light-weight wrapper around coco-ssd from the TensorFlow.js examples page.