FAQ #1: Section 3.5.2 of the evaluation plan states "A MED system processes the metadata store detecting instances of each event independently." What does this mean?
ANSWER: The results of any event detector cannot be used to influence any other event detector. For example, if event detector 1 is run on clip A, then the output of event detector 1 for clip A cannot be used to improve the performance of event detector 2 for the same clip A.
FAQ #2: Can we incorporate an event detector result from MED '11 in the metadata store (CDR)?
ANSWER: No. The intent of the CDR is to be a metadata store consisting of component detections (e.g. objects, activities) that can be re-used for Ad-Hoc event detection not the degenerate case of MED detection outputs.
FAQ #3: Can we use clips annotated for an event (Event X) as negative training examples for another event (Event Y)?
ANSWER: No. It is inappropriate to intentionally use event clips from another event to expand the negative clips for an event. However, to fill in negative clips for an event you may use all MED '11 clips, as long as they are randomly selected and not selected based on the MED '11 event label.