Image Compression and Generation using Variational Autoencoders in Python using PyTorch.
To create a machine learning project based on the Variational Autoencoder architecture. Our data comprises 60.000 characters from a dataset of fonts. Here I have trained a variational autoencoder that will be capable of compressing this character font data from 2500 dimensions down to 32 dimensions. The same model will be able to then reconstruct its original input with high fidelity. The true advantage of the variational autoencoder is its ability to create new outputs that come from distributions that closely follow its training data: we can output characters in brand new fonts.