-TalkingHeads-
A set of 4 sequences (the dataset will be enlarged soon) with top-down view, where groups of 4 people are discussing. For each person, audio data (the pitch) has been separated and stored. Audio and video are synchronized at 25 fps. For downloading and more info, go at the dataset page
-CoffeeBreak-
Two sequences taken in a coffee break situation, useful for discovering social interactions. Calibration parameters are available. For some person, tracking data, head orientation data, and the ground truth (who is socially interacting with whom) are available. For downloading and more info, go at the dataset page