File size: 3,317 Bytes
d4035c1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
-------------------------------------------------------------------------------
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.
-------------------------------------------------------------------------------
|