This code is based on Aurelien Lucchi code, SSVM. It has been adapted to compile on MACOSX with CLANG.

Things are not all set yet. Any comment/remark is welcome --> [email protected]

cd ~Code structure

• core contains the core classes
• lib contains the third party libraries used by the code

Dependencies

• opencv-dev
• libsvm (modified to be multi-thread)
• ITK 4.8.X (install with review flag ON)
• freeglut3-dev libglew1.5-dev libxmu-dev
• sudo apt-get install g++ make cmake doxygen graphviz libboost-dev libboost-graph-dev libboost-program-options-dev
• gnuplot

Windows will have to at least download cmake-gui, opencv and boost.

Provided Dependencies

• libDAI
• slic
• auxiliary ITK classes

Sources

clone with

`git clone https://github.com/elPhasmoLoco/ilastik-ssvm

Install :

1. Install the code in $LOCALHOME/src/EM/superpixels/ or $HOME/src/EM/superpixels/

2. Compile third-party libraries (linux):

3. Go to slic/build and type cmake .; make (you might have to create the build directory)

4. Go to libDAI024/build and type cmake ..; make

5. ITK : Download version 4.80.1 from the web sitei and use ccmake .. to set review flag to ON. Then build

6. Compile third-party libraries (windows):

7. Go to slic/build and type cmake-gui .. and then MSBuild.exe supervoxel.sln /p:Configuratio=Release /m

8. Go to libDAI024/build and type cmake-gui ..

9. ITK : Download version 3.20.1 from the web site and use cmake-gui to set review flag to ON. Then build.

10. Main code

mkdir build
cd build
ccmake ..
make -j

You can edit CMakeLists_common.txt or just change the flags with the ccmake interface. You might want to turn off some of the dependencies. Look at the USE_??? flags (you might need to toggle advanced mode pressing t key).

Test volumes

You can download a test volume and config file from dataset and config file. From the directory where the binaries are located, we would run:

./train sample_config.txt ./predict -c sample_config.txt -w parameter_vector0/iteration_400.txt -v 1 -g 12 ./predict -c sample_config.txt -w parameter_vector0/iteration_400.txt -v 1 -g 12 > output_predict.txt 2>&1

change the paths where images are located accordingly on the config to run the algorithm on your dataset.

Troubleshooting

On windows, when building slic/superpixels might give the error "No Target Architecture". Solve by adding the definition SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /D_AMD64_") to the CMakeLists.txt