TensorFlow.js is an extremely powerful machine learning framework. But its demonstration apps can be a little intimidating for newcomers. I wanted to make a straightforward example of a model that demonstrates the basics of TensorFlow.js. In this article walks through a simple classification model which solves for XOR.
Google’s Cardboard viewer standard has no standard control scheme. This is what led me to write nod.js, a simple gesture based event system for Google Cardboard enabled web apps. It uses device accelerometers to detect a sharp motion in one of four directions: up, down, left and right. With nod.js you can implement actions such as next, previous, confirm and cancel without the need for an external controller.
A recent project I worked on involved a thermal camera. At first it wasn’t clear whether or not we would be able to wire it into a web app. I decided to create a few alternative prototypes to fall back on, one of which involved motion detection. Anything moving is likely to produce heat. The effect works by capturing two frames, 3 seconds apart, from the live video feed. Even someone holding still to pose for the camera moves a little. Cheating, for sure, but the effect works really well.