Subject 32 was chosen for testing CERT builds because that video contains examples of all of the main AUs. Shell scripts in the WinBuildAndTest directory perform all of the following steps:
    Build the release version of CERT
        The buildProject.sh script uses Microsoft Visual Studio to do this step.
    Run CERT on the Subject 32 frames.
        The command line would be: ./CERTDemo.exe -t -gc ../../GKR/D006_S032_GKR.xml ../../D006_S032
        The -t makes CERT output simple tab delimited data, one line per frame, with the labels for each field on a line just before the first frame.
        The "-gc <filename>" options tells the CKF plugin where to find the subject specific neutral model.
        The last argument is the path to the frames for subject 32.
    Clean up the output so that it can be read into Matlab.
    Use Matlab to calculate the ROC values for all 11 main AUs plus four MLR outputs.
    All outputs from the build and run plus the ROC outputs are saved to files and added to the CERT release zip file.
The CERT_Test.sh script starts the test. The script depends on the Subject 32 frames being in a directory called D006_S032 at the same directory level as the directory that contains the scripts.

NRL Test Set (D007 subjects)
All of the scripts and data files for this test are in the WinBuildAndTest/NRL_Test directory.
The NRL_Test.sh script runs a large number of subjects. Run times can be over 24 hours.
Output for each subject goes into it's own file. All files are then cleaned up so they can be read directly into Matlab, where ROCs are calculated and written to a file in the same directory with the raw output.
The NRL_Test.sh script assumes that all subject directories on at the top level of the E: drive. The BASE script variable can be changed to point to a different location.


A more complete description of how to train and use GKR.

Gather images of a particular subject that have a neutral expression. That is, those frames where none of the AUs of interest are present.
Run all of these images through CERT with the addition of the following command line arguments:
-gt <path to save model file>
The '-gt' (i.e. 'GKR Training') option tells the GKR portion of the CKF plugin to assume that the given images are neutrals and to use them to train a model for that particular person.
The saved model file can then be used as a baseline when testing images from that subject. The command line arguments would then be:
-gc <path to saved model file>