22 lines
994 B
Plaintext
22 lines
994 B
Plaintext
This software implements two algorithms for minimizing energy functions of the form
|
|
|
|
E(x) = \sum_i D_i(x_i) + \sum_ij V_ij(x_i,x_j)
|
|
where x_i are discrete variables.
|
|
|
|
The two algorithms are max-product belief propagation (BP, Pearl'88) and
|
|
sequential tree-reweighted max-product message passing (TRW-S, Kolmogorov'05).
|
|
|
|
For example usage look at one of the type*.h files
|
|
(typeBinary.h, typeBinaryFast.h, typePotts.h, typeGeneral.h,
|
|
typeTruncatedLinear.h, typeTruncatedQuadratic.h, typeTruncatedLinear2D.h, typeTruncatedQuadratic2D.h).
|
|
If your energy function does not belong to the classes defined
|
|
in these files but terms V_ij allow fast distance transforms, then
|
|
it should be possible to extend the algorithms to your functions
|
|
using files type*.h as examples.
|
|
|
|
Written by Vladimir Kolmogorov (vnk@microsoft.com), 2005.
|
|
Tested under Microsoft Visual Studio .NET (Windows)
|
|
and GNU c++ compiler version 2.96 (Red Hat Linux 7.1).
|
|
|
|
(c) Microsoft Corporation. All rights reserved.
|