Recent developments on video indexing have shown that this indexing is only valuable if done at a sufficient semantic level. On the other hand, it is known that such a semantic level cannot be reached without drastically constraining the domain. Hence, the indexer is then left with essentially two strategies. Either giving up on corpus genericity and therefore provide a specific automated proessing of the content. Alternatively, a similar semantic level may be attained within a generic corpus at the cost of required user assistance and therefore privilege interaction.
 Here, we show how we have mixed both strategies in Vicode, our video indexing platform with using salient content as a path to specificity while remaining flexible to online learning based on user relevance feedback in a QBE search paradigm.