COGNIMUSE is a multimodal video database annotated with saliency, events, semantics and emotion with application to summarization. It can be used for training and evaluation of event detection and summarization algorithms, for classification and recognition of audio-visual and cross-media events, as well as for emotion tracking.