Semantic video labeling by developmental visual agents