ZipNet

ZipNet is a Work-In-Progress for the course project of Data Compression With Deep Probabilistic Models offered by Robert Bamler at the University of Tübingen.

We want to provide a fully functional Neural Image-Encoder and Decoder on the Web. The goal is to achieve a superior compression rate over classical codes (JPEG, …) while retaining acceptable performance for a Web Application (compression in a few seconds). This results in a maximally portable application that could help Neural Compression Codecs achieve a higher adoption rate. For the implementation, we leverage the new, performant WebASM standard along with a simple, flexible, yet powerful model architecture.

You can track our progress in the linked GitHub repository and on our website.