-------------------------------------------------------------------------------
Version 1.2.0
- Thanks to the following individuals, especially John Kaufhold, whose
  patient use of the version 1.1.0 benchmark and extremely helpful comments 
  and suggestions led to this hopefully much improved release:
        John Kaufhold (GE Research)
        Stefano Casadei
        Robert O'Callaghan (University of Bristol)
        Stephen Zheng
- moved thinning operations in boundaryPR.m and boundaryPRfast.m after
  the thresholding operation.  This way, we can benchmark boundary maps
  that have not been thinned first (though it is always best for users
  to do whatever nonmax suppression they can beforehand)
        ** This has the effect of improving the scores of algorithms
           on the benchmark, as thinning before thresholding tends
           to remove more boundary pixels.
- removed thinning operation from boundaryBench.m when upsampling a boundary
  map that is smaller than the segmentations, since it is unnecessary
- added best F measure points to all graphs
- boundaryBenchHtml.m and boundaryBenchGraphsMulti.m scripts are more 
  robust w.r.t. half-completed algorithm directories (they skip them)
  and missing human pb directories
- [internal] setup.pl script now generates iids_{train,test}.txt in
  BSDS directory
- iids_{train,test}.txt now included in BSDS-images.tgz tarball
- made sure all calls to correspondPixels pass double arrays instead
  of logical arrays, so that we don't get those warnings about accessing
  a logical array with mxGetPr (instead of mxGetLogicals)
- in Util/Matrix.{hh,cc}:
        - disclude nextpow2 methods by default, since they depend
          on the system header ieeefp.h (which doesn't exist on some
          systems)
        - made helper scalar math functions static
        - commented out log2() and trunc(), since they now appear to be
          in libc.  If your libc doesn't have these, then you need to
          uncomment them.
- added -DNOBLAS to compilation of Util/Matrix.cc so that one does not
  need a BLAS library to use the Matrix class.  None of my code uses
  this class to do matrix multiplcation.  If it did, then this would
  impact the performance of such code.
- removed unused runtest directive from Benchmark/GNUmakefile
- added all Benchmark/*.m files to install directive in Benchmark/GNUmakefile
- fixed non-critical idiag bug in Benchmark/correspondPixels.cc
- fixed declaration of nomem_msg in CSA/csa.{hh,cc}
- removed extern "C" from Util/isum.c; mex doesn't seem to need it, and it
  was causing build problems for some
- bugfix: added fclose(fid) to end of writeSeg.m
-------------------------------------------------------------------------------
Version 1.1.0
- Created single feature gradient detectors {bg,cg,tg}.
- Retrained {bg,cg,tg,bgtg,cgtg} detectors using either gray or color
segmentations, depending on whether the detector is to be benchmarked
as a gray or color algorithm.  The bgtg detector was trained with both
to have a comparison point.
-------------------------------------------------------------------------------
Version 1.0.0 
- Initial public release.
-------------------------------------------------------------------------------