Installation

The current package is just a set of wrapper classes for the CSU facerec2010 module, which is contained in the CSU Face Recognition Resources, where you need to download the Baseline 2011 Algorithms.

Patching the CSU Face Recognition Resources

To be compatible with bob.bio, the CSU toolkit needs to be patched. If you haven’t patched it yet, please follow the set of instructions:

1. Download the bob.bio.csu package from our PyPI page and extract it into a directory of your choice.

2. Generate the binaries of this package without the CSU toolkit. We provide a special buildout configuration file for that:

   $ python bootstrap.py
   $ ./bin/buildout -c buildout-before-patch.cfg
   

   This will disable the CSU code for a while.

3. Patch the CSU toolkit by calling:

   $ ./bin/patch_CSU.py [PATH_TO_YOUR_CSU_COPY]
   

   If you get any error message, the sources of the CSU might have changed (the latest test was done in December 2012). Please file a bug report in our GitHub page to inform us so that we can provide a new patch.

4. Update the CSU toolkit path in the buildout.cfg file by setting the csu-dir variable via replacing the [PATH_TO_YOUR_CSU_COPY] with your actual directory:

csu-dir = /path/to/your/csu/copy

and re-generate the binaries, this time including the CSU toolkit:

.. code-block:: sh

$ bin/buildout

or simply re-generate the binaries with the option:

$ bin/buildout buildout:csu-dir=/path/to/your/csu/copy

Note

When you are working at Idiap, you might get a pre-patched version of the CSU Face Recognition Resources.

Warning

After patching the CSU toolkit, the original experiments of the CSU toolkit will not work any more! Maybe it is a good idea to make a save-copy of your CSU copy before applying the patch.

Verifying your Installation

After the CSU toolkit is patched, please verify that the installation works as expected. For this, please run our test environment by calling:

$ bin/nosetests -vs

Please assure that all 6 tests pass.

Running CSU experiments with bob.bio

The easiest way to run any experiment with the CSU tools is to use bob.bio directly, using any of the databases from bob.bio.face. After running the command lines above, the CSU tools should be registered as Resources, i.e., they are listed in the:

$ ./bin/resources.py

and can be used on as a command line parameter like:

$ ./bin/verify.py --preprocessor lda-ir --extractor lda-ir --algorithm lda-ir --database gbu ...

Additionally, now two new baseline experiments lrpca and lda-ir can be run in using the ./bin/baselines.py script, see Executing Baseline Algorithms.

Please check the bob.bio Documentation on more details on how to run face recognition experiments using the above mentioned two scripts.

One example on how to compare the CSU algorithms to other state-of-the-art algorithms using the FaceRecLib (on the base of which bob.bio is originally build) is given in our paper:

@inproceedings{Guenther_BeFIT2012,
       author = {G{\"u}nther, Manuel AND Wallace, Roy AND Marcel, S{\'e}bastien},
       editor = {Fusiello, Andrea AND Murino, Vittorio AND Cucchiara, Rita},
     keywords = {Biometrics, Face Recognition, Open Source, Reproducible Research},
        month = oct,
        title = {An Open Source Framework for Standardized Comparisons of Face Recognition Algorithms},
    booktitle = {Computer Vision - ECCV 2012. Workshops and Demonstrations},
       series = {Lecture Notes in Computer Science},
       volume = {7585},
         year = {2012},
        pages = {547-556},
    publisher = {Springer Berlin},
     location = {Heidelberg},
          url = {http://publications.idiap.ch/downloads/papers/2012/Gunther_BEFIT2012_2012.pdf}
}

The source code for this paper, which actually uses the FaceRecLib, can be found under http://pypi.python.org/pypi/xfacereclib.paper.BeFIT2012.

Note

The source code for http://pypi.python.org/pypi/xfacereclib.paper.BeFIT2012 depends on an older version of Bob and an old version of this package, and is not (yet) ported to the new Bob version 2.0.