ANDpakk2 is an EXEcutable compression tool allowing
you to get better compression ratios of your intros,
demos or any other kind of software.
This project is a part of research work of the author
for internal usage and 64k intro experiments.
ANDpakk2 is based on idea of adaptive statistical data
compression based on context modeling and prediction by
model mixing and prediction by partial matching on
neural networks.
All the data is getting processed by filters,
predicted in different context models (by neural
networks) which are getting mixed and predicted by
another neural networks again. Prediction is done
per bit with four-byte context history and coded by
arithmetic coder at the final stage.
I've got very impressive compression ratios making
Zoom3 64k intro compressed down to 45760 bytes after
some specific optimizations of the data structures.
It also shows much much better compression ratios than
any other general purpose EXE/data compressor.
|