Build App Prototypes using the Power of Deep Learning Based AI
What if making an app prototype is as easy as selecting some images?
We have tried to solve exactly this based on a easy to use process, where the user selects the type of the application he would like to begin with, eg: Social or Travel app and then chooses a couple of UI screens and a basic colour palette. The webapp then makes predictions based on the selected screens and the colour palette to suggest them a UI prototype which is visually appealing and readily presentable
We use a rich dataset having more than 70,000 UI Screens from various apps on the play store. The Dataset can be found here : http://interactionmining.org/rico
We use the trained vectors from a Deep Convolutional Autoencoder to find semantically similar images to the ones selected by the users.
We also have custom classifiers for different Views for Android Apps - ListView, CardView and GridView. These provide the parameters like optimal size, position and colour for the UI screens. We use a custom Convolutional Neural Network roughly based on the VGG architechture.
This information is then fed to the templating system which builds a visually appealing prototype for the user
There is a lot of possibility for extension, and we have just scratched the surface of the possibilities in this Hackathon.