Change Log

Made test slightly more robust.

clarified spec

Clarified spec regarding infinite factor potentials and flow capacities.

Added tests to make sure the graph cut stuff works with
infinite weights on some edges.

All I did in this change was rename a bunch of reinterpret_casts
to static_casts whenever it was possilbe to do so.

A minor change to avoid warnings about violation of string aliasing rules.

Changed to avoid warnings in valgrind about invalid reads.
The manpages seemed to indicate that the previous code was
correct so I'm not sure if this was a bug or not.  It's been
in dlib for years without causing trouble though.

Added unit tests for the grid version of find_max_factor_graph_potts().

- Made scale_by() work on dlib::matrix objects.
 - Added an add() and subtract() that works on
   sparse and dense vectors.

Changed the is_signed_type and is_unsigned_type templates to work
with floating point types in addition to integer types.

Clarified some specs, added a missing requirement to sum_filter(),
and added overloaded versions of a few image processing functions
to make doing in-place operations slightly more convenient.

A minor change to avoid a compiler warning from gcc 4.4.

Fixed a bug which caused you to get a compiler error if you tried
to call dot() on two 1x1 matrices which were statically dimensioned.

clarified spec

Clarified spec and added more tests.

fixed spelling error in comment.

Switched all the graph labeling stuff to use bool as a node label
rather than the node_label type from the min_cut object.  This
should make the interface much less confusing.

Added some tests to make sure the recent addition to svm_c_linear_trainer
works right.

Added the ability to learn only non-negative weights to the svm_c_linear_trainer.

Clarified spec

clarified spec

minor cleanup

Added another overload of find_max_factor_graph_potts() that works on
graphs that are regular grids.

Added missing ;

clarified spec

Fixed typo in spec

clarified spec

Added some compile time asserts.

Changed the spec to say that array2d objects don't put padding at the
ends of the rows.

Changed the SINK_CUT and FREE_NODE constants so that when we use graph
cuts on an image the result is a displayable binary image rather than
an image which is super dark and hard to view.

Clarified spec

Fixed typo in spec

Optimized this code a little.

Improved aserts

Fixed typo in spec

Just moved some files and #includes around a little.

Fixed a bug in find_max_factor_graph_viterbi() which sometimes
occurred when the model order was larger than the number of variables.

Updated the interface to the structural_graph_labeling_trainer so the user
can set the per class loss to whatever they want.

Fixed minor bug in the set_loss_on_negative_class() function.

Added missing comments.

Added the ability for the user to set the per class loss.

Slightly improved messages from some asserts.

removed cruft

Added a compile time assert to hopefully simplify possible error messages.

Added some more tests

Just added some casts to avoid compiler warnings.

Added missing assert.

Simplified the code a little.

Added the += and -= operators to the set_subm(), set_rowm(), and set_colm()
functions.

Fixed out of date docs

Added some unit tests for sparse_to_dense()

- Added make_sparse_vector()
 - Refined the sparse_to_dense() routines a little.

Added unit tests for the graph labeling stuff.

Changed graph_labeler requirements to allow for zero sized sparse vectors.

Fixed bug in is_graph_labeling_problem().  It didn't allow
for zero sized sparse vectors.

Fixed a bug in sparse_to_dense() which I just introduced a few
commits ago.

Added more overloads of sparse_to_dense().  Now it works on single
vectors of both sparse and dense type.

Moved sparse_to_dense() into the file with all the other sparse
vector functions.

Made this file #include the new graph cut learning tools.

Filled out this spec file

updated #includes

Filled out spec file

removed cruft

Clarified spec

Added asserts

Added missing requirement to the dot() function for matrices.

improved spec

Filled out spec file.

Fixed a very minor typo in the spec.

Apparently, XKeycodeToKeysym() is deprecated.  Since clang++ was
giving warnings about this I switched to the replacement function
XkbKeycodeToKeysym().

Gave the sparse vector tools their own #include file.

Removed the dlib::sparse_vector namespace.  I put everything from this
namespace into the normal dlib:: namespace so that code which works
with both sparse and dense vectors is more cohesive.

Added overloads for dot() so you can dot() a std::vector with a std::map.

Backed out changeset: 83039c25f3cc
This change runs into a bug in visual studio 2005 that I can't seem
to find a way to avoid.  Might try it again later, but for now, I'm backing
it out.

Made the sparse vector dot() more flexible.

added a comment.

Fixed the code so it works with sparse vectors.

Filled out the implementation of is_graph_labeling_problem().

Added unit tests for new sparse vector routines.

Added functions for finding the min and max elements of a sparse vector.

Made these functions work even if one the labels never shows up
in the data.

Filled out spec file

Added validation functions for graph labeling problems.

Added the copy_graph() routine.

updated a spec

Renamed a function.

removed cruft

Added initial version of graph labeling trainer.

minor cleanup

switched to non-checking version.

Added initial version of graph_labeler function object.

Just renamed some things

just renamed some files

Refined function contract a little.

Added initial sketch of a potts learning problem definition.

Added initial version of structural_svm_potts_problem object.

Removed unneeded code.

Changed the oca interface to allow you to specify that a range of w
elements should be non-negative rather than just being able to say
all or none of them are non-negative.

Added missing asserts and requires clauses

Added some more unit tests.  Also fixed a bug in the potts_model_score() routine.

clarified spec

clarified spec and fixed typos.

Added an overload of find_max_factor_graph_potts() which takes a regular
undirected graph rather than the special purpose potts_problem type
object.

Added another print statement to an assert

Simplified the potts model interface a little.

minor change to avoid compiler error in visual studio.

Added a comment about when this routine gives the exact MAP vs.
just an approximate answer.

clarified spec

clarified spec

Improved spec organization a little.

Added unit tests for the new graph cuts tools.

Added graph cut tools.  This is the new min_cut algorithm and find_max_factor_graph_potts()
routine.

Fixed typo in inclusion guard.

Added the graph_has_symmetric_edges() routine.

Clarified spec

Fixed some bugs in toMat() which prevented it from compiling sometimes.

Added the toMat() routine for converting from a dlib style image to an
OpenCV cv::Mat image.

Updated the cv_image object so it works with cv::Mat as well as IplImage.

Added comments describing the bridge protocol.

Added another overload of edge() for directed graphs.

Fixed typo in spec

Added some unit tests for the oca object.

Minor change to avoid a warning from gcc.

Added an optional non-negativity constraint on w to the oca optimizer.

Added unit tests for the new quadratic program solver.

Added another quadratic program solver.

fixed grammar

Added lowerbound() and upperbound() routines.