The scripts and data files in this folder are used to calculate the Area Under the ROC for selected subjects and frames of from the D007 data set.


frames.txt is a Matlab loadable list of which frames should be used from each subject. The fields are: Subject number, 1 for primary interview and 2 for secondary screening, start frame, end frame.

Labels2.mat contains the labels for all AUs for all frames for all subjects.

The clean.sh shell script removes text from the CERT output files so they can be loaded into Matlab. The "-4" command line option will clean files output by CERT 4.0. The default is to assume files from CERT 4.3.

runfiles.m contains an example of how the other Matlab scripts can be called to calculate ROCs for the entire test set.

NRL_ROC.m takes a file name, a subject number as a string, either a '1' or a '2' (for primary vs secondary screening) and an optional boolean flag telling the script to print the results out or just return them. Results are for the non-GKR AU outputs and the GKR outputs.

NRL_ROC_v4.m is similar to NRL_ROC.m except it works on CERT 4.0 output.

ROC_area.m is used by both the above scripts to actually calculate the ROCs.

getFirstLastFrames.m gets the starting and ending frame numbers from the frames.txt file.