Friday, October 17, 2014

Media content analysis demo



We are working on movies to automatically extract a variety of information. In the demo video, you'll see a clip from the movie Good Deeds, where we detect characters' faces, identify their gender (notice the color of the bounding box), detect whether they are speaking or not (a dot appears when a person speaks), and even recognize one speaker from another.

Emotion In Music

MICA participated in the MediaEval2014 challenge in two subtasks:

Emotion In Music
Kumar et al., "Affective feature design and predicting continuous affective dimensions from music", MediaEval, Nov 2014.

Crowdsourcing for Social Media.
Kumar and Narayanan, "Detection of musical event drop from crowdsourced annotations using a noisy channel model", MediaEval, Nov 2014.