kvendrik / polaris-whiteboarder

[Experiment] Creates simple Polaris prototypes from wireframes drawn on a whiteboard

Polaris Whiteboarder

This experiment creates simple Polaris prototypes from wireframes drawn on a whiteboard. It has many flaws but was fun to work on.

See it in action

How it works

1. Training data is uploaded to IBM Cloud Annotations and labeled so the machine knows what components are visible in the images.
2. We use a Watson Machine Learning instance to train a model using this training data.
3. We download this newly trained model and store it with the front-end (this repository).
4. The front-end then uses Tensorflow.js to (client-side) detect prototype-friendly Polaris components in a given image (utility code is highly derived from IBM's object detection library (the only reason it was taken apart was to learn more about Tensorflow)).

This flow is based on IBM's object detection walkthrough.