###########################################################################
# # # HCM - software for learning hierarchial compositional model # # #
###########################################################################

1. Introduction.

This software library implements the hierarchial compositional model learning algorithm
described in

    "Unsupervised Learning of Dictionaries of Hierarchical Compositional Models."
    Jifeng Dai, Yi Hong, Wenze Hu, Song-Chun Zhu, and Ying Nian Wu. CVPR 2014

----------------------------------------------------------------------

Tested under windows 7 x64, matlab R2011a, Visual 2008 C++
compiler.

##################################################################

2. License & disclaimer.

    This software can be used for research purposes only.

##################################################################

3. Example usage.

This package contains a subset of the image representation, domain transfer and cosegmentation examples due to space limit. The basic usage is as
follows:

3.1 For image representation:

1) Start matlab and set the current folder to 'net', 'bike' 'face' or 'caltech'. 

2) Run the 'StartFromHere.m' script to perform hierarchial compositional template learning. The input images are stored
in folder './positiveImage'.

3) After the program finishes, check the './document' folder
for the results. The results are shown by html files generated
by the program automatically.

4) Intermediate results are stored in the './output' folder.

3.2 For domain transfer:

1) Download the four domain datasets including caltech 256, amazon, DSLR, and webcam. Put them in 'FourDomain\dataset'

2) Start matlab and set the current folder to 'FourDomain'. It is recommended to start matlab parallel computing
environment.

3) Run the 'fourDomainExpV3.m' script to perform domain transfer. The resulting confusion matrix is stored in the returned 'confMat'.

3.3 For cosegmentation:

The cosegmentation result of the proposed approach is presented in folder 'ImageNet'.

##################################################################

4. Notice

We utilized codes provided by other researchers which are
publicly available, e.g. FFT. Please check the
licence of them if you want to make use of this code.


///////////////////////////////////////////////////
