"""
Name: readme.txt
Author: Yuxiang LI (li.yuxiang.nj@gmail.com)
Data: 25/03/2016

## Title: a Python version of Burger's MLP denoiser

## Description:

This program is based on Burger's paper: Can plain Neural Networks compete with BM3D?. 
It is implemented with Keras, theano and Scipy. To run this program, you need to download 
the network weights which can be found on author's website (CVPR one). After that, use 
load_data.py to generate Keras model and run denoise.py to see the result.

If you don't want to install Keras and theano which is usually not evident, you can add
"--cpu" to the argument. In this case, only the .mat file is required.

The result is slightly worse than the Matlab version, it may be due to the random 
number generation. To get a better result, you are invited to use the full version which 
contains a machine learner based on Burger's configuration and deals with this difference. 
The full version will be made public on my Github (lyx-x) as soon as I can.

## Tips for installing Keras and theano:

1. First install Keras from the Github
2. Then install the latest theano from the Github

## Demo: python denoise.py [--cpu]
"""

