-------------------------------------------------------------------------------

TO BENCHMARK A BOUNDARY DETECTION ALGORITHM

(1) Download and untar the benchmark images and human segmentations in
the same location.  This will create a BSDS300 directory.

(2) Run your algorithm on the 100 test images.  You may use the 200
training images and associated human segmentations to calibrate your
algorithm.  The output of your algorithm should be a soft boundary
map, preferably after nonmax suppression, stored in a BMP image file.

(3) Download and untar the boundary detection benchmark code.
Edit the matlab function Dataset/bsdsRoot.m so that it points to
the BSDS300 directory created in the previous step.  Now run
'gmake install' from the root code directory.

(4) Create a web-accessible directory to contain the benchmark results.
Call it 'BENCH'.

(5) Download humanpb.tgz, and untar it from the 'BENCH'.
This should create two directories 'BENCH/gray/human' and
'BENCH/color/human', each containing many files.

(6) The algorithm to be benchmarked works either on grayscale images ('gray')
or color images ('color').  Call this choice between 'gray' and 'color'
'PRES'. Create directory for the algorithm 'BENCH/PRES/ALG'.  

(7) Put the algorithm's pb files in the directory 'BENCH/PRES/ALG' as
'<iid>.bmp', where <iid> is the corel image ID for each image.

(8) Create a 'BENCH/PRES/ALG/name.txt' file containing a single line of
text identifying the algorithm.

(9) Create a 'BENCH/PRES/ALG/about.html' file in the algorithm directory
containing a short description of the algorithm.

(10) Start up matlab.  Add the lib/matlab directory to your matlab path.
This directory was created when you build the benchmark code in step 3.

---> Either follow the remaining directions, or use the benchNewAlg.m
     script.

(11) Run boundaryBench('BENCH/PRES/ALG','PRES').  This will compute
precision/recall and put the results in text files in the ALG
directory.  This takes 120 minutes on my 3GHz P-IV.

(12) Run boundaryBenchGraphs('BENCH/PRES/ALG').  This will create
precision/recall graphs in the ALG directory.  

--> If you have other algorithms to benchmark, repeat steps 6-12 for
    each algorithm.

(13) Run boundaryBenchGraphsMulti('BENCH').  This will create graphs
comparing pairs of algorithms.

(14) Run boundaryBenchHtml('BENCH').  This will generate all the web
pages for the benchmark.

-------------------------------------------------------------------------------