A WebDNN implementation of a variational autoencoder trained on the shoe sector of Fashion-MNIST
Hosted at https://petosegan.github.io/shoespace-online/
A variational autoencoder is a neural network that is trained to embed high dimensional data (such as images) into a lower-dimensional space. This one has been trained to encode 28x28 black and white pictures of shoes into a two-dimensional space.
The network was trained on the shoe images from the Fashion-MNIST using Keras. The trained weights were converted to WebDNN with a WebGL background for performing inference in the browser.