*********
As of November 28 2013, the optimization toolbox is DISCONTINUED.
The author has permanently left academic research. He may react to e-mails
*********


The optimization toolbox is primarily a library, i.e. it does not provide
executable programs (although in the future there will be sample programs).

At the moment, it covers three areas:

1. MESSAGE PASSING for solving discrete labeling problems: here you can choose
between belief propagation and various methods to solve linear programming
relaxations. For the latter, you can choose between relaxations with singleton
separators (TRWS, MPLP, MSD, Subgradient methods) and with pairwise separators
(TRWS, MSD, Subgradient methods).   
 The classes provide various forms of higher order terms, including some 
efficiently handable terms of very high order (1-of-N, cardinality, integer
linear constraints).
 These classes were developed by the author, but in frequent correspondence
with Vladimir Kolmogorov. 

Associated files: factor*, sep*


2. SUBMODULAR FUNCTIONS: here there is code to check a function for
submodularity and to convert a submodular function into a form that can be
optimized via graph mincut. The latter relies on code from Vladimir
Kolmogorov, which you will have to download.

Associated files: submod*, check_submodularity.*, function.*


3. NBEST SEARCH in graphs. This is particularly useful for Natural Language
Processing.

Associated files: nbest.*


A sample Makefile has been provided. It compiles all source files that do NOT
depend on thirdparty code into a library.