Workshop Program Updated
The automated analysis of social interactions has attracted increasing interest from the computer vision community. A great deal of meaning in social interactions is subtly encoded in the posture, movement, expressions, tone of voice, etc. of the participants. Extraction of meaning from such subtle cues requires novel analysis methods that go beyond standard techniques for the classification of actions and activities. The goal of this workshop is to define and explore new challenges and approaches for decoding subtle human behavior in social interactions. Relevant topics for the workshop include but are by no means limited to:
The workshop invites interested participants to submit research papers (8 pages) presenting original research. In addition to research papers, the committee will also accept extended abstracts of ongoing or published work (2 pages). Submitted papers should follow the ICCV 2013 paper format, used for the main conference. All submitted papers will receive double blind reviews. Authors may choose whether accepted papers appear in the proceedings. We expect accepted papers will be presented either orally or as a poster, depending as resources allow. Please follow the CMT system to submit your manuscript.
The workshop will also introduce the Multimodal Dyadic Behavior Dataset (MMDB), which will serve as the basis for a new competition. MMDB is the result of a major, three year collection effort by a team investigating the behavioral basis of development in young children. MMDB is by far the largest publicly available dataset of multimodal (video, audio, and physiological) recordings of the social and communicative behavior of toddlers. We will explain how to manage issues of privacy and human subject permissions in using this dataset. MMDB contains material wholly unlike current activity recognition datasets. For example, it has 160 sessions of 3-5 minute semi-structured play interactions between a trained adult examiner and a child, with 5 rated activities and 18 non-verbal communication behaviors. These sessions are viewed from multiple cameras, including a range sensor. There is extensive markup by domain experts indicating particular kinds of important behaviors and interactions. MMDB supports many different types of analysis tasks, including: