Executing Vulnerability Analysis Experiment

It is desirable to conduct this experiment at last (or after single channel PAD experiments) so that it can re-use the extracted features of CNN+LR method on VIS channel.

However, the PAD experiments do not use the pics data from XCSMAD. These pictures should be processed as follows:

1. Run the script process_pics.py from the package to obtain the extracted features from pictures data.

2. Now features of complete XCSMAD data have been extratced, but in different directories (one for videos, and another for pics). Bob requires the data to be present in the same directory. Therefore, you should create a dummy extracted directory which provides links to the extratced directories of videos and pics. This can be accomplished by script link_verification_data.py in the package.

3. Subsequently the verification experiment can be run using verify.py command.

4. The matching scores should be split into licit and spoof scores to analyse its vulnerability. The script split_vuln_scores.py will generate two separate score files in the same score folder— which should be provided to the bob command to generate histogram of scores.

All the necessary commands are generated through the provided script.