Evolve

Using genetic algorithms for universal image compression.

Motivated Design

Naturally motivated genetic algorithm design from numerous testing iterations

Interprability

Well-defined metrics and genetic algorithms

Quality Engineering

We pride ourselves in the quality, sturdiness, and usability of our final product

Design

We break down images into codeblocks that we aim to evolve over training to obtain a set of general codeblocks - a codevector, that any given image could be represented in.

By stochastically crossing and mutation different codevectors based on their fitness - performance in encoding the training set of images - we evolve a general codevector that can be used to represent any given image with a constant compression rate.

App
7 Team Members
8 CSUA GPUs
0.9 SSIM Loss
1234567890 Codeblocks Tested

Results

Take a look at some of our compression results

Original on the left, Compressed on the right