This paper is about data collection for face recognition. One idea was to use weaker classifiers to rank the data presented to the annotators.