**** A Grid Conditional Random Field Toolkit for Matlab ***


                    Jerod Weinman
                    jerod@acm.org
                     7 May 2009

The GCRF toolkit handles Conditional Random Fields in a grid topology
assuming potentials at nodes and edges are the same everywhere in the
grid (though vertical and horizontal edge weights may be distinct or
tied). The structure of the object-oriented model is as folows. Each
@gcrf object stores the weights for its potentials. Data for training
and prediction uses an array of general grid graph feature struct with
observation features for each node and edge in a specific grid (sizes
may be variable). The weights may be learned by several variants of
belief propagation. (See examples in the README below.)

This code was designed, not to be optimized, but to be correct first,
though we may not have succeeded at even that in places. While I am
likely unable to provide support, bug reports are welcome, as are
small patches for optimizations. Much of the code was written before
the author was aware of cellfun or before bsxfun was introduced to
Matlab, and these are likely points for useful optimizations.


INSTALL

# Download gcrf and lbfgs
wget http://www.cs.grinnell.edu/~weinman/code/gcrf.tar
wget http://www.cs.grinnell.edu/~weinman/code/lbfgs.tar

# Make destination and unpack
mkdir gcrf
tar xf gcrf.tar -C gcrf
tar xf lbfgs.tar -C gcrf

# Start matlab
matlab

% Add commands to path
>> addpath(genpath('gcrf'))

% Read the demo
>> edit recognition/gcrf/README.m

% Compile a BP accelerator
>> cd inference/mex
>> mex calcMsgProd.c

% Run the demo
>> cd recognition/gcrf
>> README