Pokemon generation and interpolation

This project uses a Variational Autoencoder to map existing Pokemon images to a latent space, and generate new images from points in the latent space. By sampling the latent space it is possible to generate new Pokemon images, and by interpolating between the latent space vectors of 2 Pokemon, we can interpolate between the Pokemon themselves.

In addition to the Colab notebook and Github links, a demo app is also partially functional. The app could not be completed due to compatibility issues between TenorFlow's Pyhton and JavaScript versions.