INSTALL

#### OSX INSTALL #####
On OSX, octave does not play nicely with C++ because it is compiled with the macports compiler instead of the stock mac gcc. So, if you
try to link to C++ libraries (that use a mix of header and compiled
logic), you start getting segfaults in weird places.

However, you can run the algorithms in the MVP as octave-only. To do
so, you must build the files in octmvp. The following assumes you have
octave installed through macports (must be >3.6, right now is called octave-devel)

(Starting in the root project dir):

$ mkdir build2
$ cd build2
$ cmake ../octmvp
$ make

(to run tests, run make test, as usual)

After this is completed, you can run the algorithms in src/mvp/Algorithm
on a matlab job file. Grab 2682_1937_12.job.mat from the example mvp data. 

(in the examples/mvprun dir):

$ octave
$ start_octmvp
$ mvprun_tile
$ imagesc(result)

To mvprun_tile instead. If you have the C++ mvp compiled
you could have run start_mvp (instead of start_octmvp), and it would have run
the algorithms using the C++ infastructure instead.

For some reason, the MVP runs super fast (by an order of 10) on OSX as compared 
to my linux machine... Could it be that the "accelerate" framework speeds it up? 

######################

###### Linux Instructions ############

#
# Configure the MVP
#

mkdir build
export VW_ROOT=<path_to_vw_build_dir>
ccmake ..
[Press 'c' to configure]
[Select option 'BINARY_DIR_INSTALL']
[Optionally select option 'BUILD_TESTS']
[Press 'c' to reconfigure]
[Press 'g' to generate]

#
# Compile and install the MVP
#

make -j4 && make install

#
# Test the MVP
#

cd MultipleViewPipeline/examples/synth
../../build/usr/bin/mvp
../../build/usr/bin/mvp --use-octave true