The Tagging Task
Requires participants to automatically assign tags to videos using features derived from visual, textual or audio/speech content of the videos.
Very interested: 10 Sounds interesting: 15 Need more info: 2 Not for me: 6
The Affect Task
Requires participants to identify videos that viewers report are boring.
Very interested: 7 Sounds interesting: 13 Need more info: 3 Not for me: 10
The Video Passage Retrieval Task
Involves the identification of relevant jump-in points in video given a set of queries based on the combination of modalities (audio, speech, visual, metadata)
Very interested: 7 Sounds interesting: 12 Need more info: 5 Not for me: 9
The Recommendation Task
Involves predicting the age-level appropriateness of videos using spoken, visual, and audio content as well as accompanying metadata
Very interested: 7 Sounds interesting: 8 Need more info: 5 Not for me: 13
The Appeal Task
Requires participants to predict the viewer-appeal of videos using spoken, visual, and audio content as well as accompanying metadata
Very interested: 7 Sounds interesting: 8 Need more info: 5 Not for me: 13
The Linking Task
Involves linking a video segment to Wikipedia material on the same subject.
Very interested: 5 Sounds interesting: 16 Need more info: 4 Not for me: 8
The Duplicate Detection Task
Requires participants to identify duplicate video material
Very interested: 4 Sounds interesting: 9 Need more info: 4 Not for me: 16
The Summarization Task
Requires participants to create “MediaClouds,” visual summaries of the contents of a video to facilitate quick browsing through search results or within videos.
Very interested: 5 Sounds interesting: 16 Need more info: 4 Not for me: 8
A selection of comments:
(On the Tagging Task) I would be interested in the more specific task of tagging people...
(On the Tagging Task) Isn’t this what they do in TRECVid? (Organizers note: No, the tags reflect aboutness or apropriateness -- we are going beyond TRECVid, which has traditionally related relevance exclusively to what is depicted in the visual channel.)
(On the Tagging Task) Will there be a development set? (Organizers note: Development set yes. Training set no. Participants collect their own data if they chose to use a machine learning approach)
(On Passage Retrieval): Getting the user to the right story sounds very interesting...
(On the Affect Task): What about appeal in different genres? And what about genre detection?
(On the Affect Task): Groundtruth is difficult to create. (Organizers note: Yup! We’re working on rising to the challenge.)
(On all tasks) Participation depends on time and budget constraints