Personal tools
You are here: Home Dataset Idiap Poster Data

Idiap Poster Data

 

 

Data Description:

00001005.jpgIdiap Poster Data is video dataset. The videos were shot during a real social event of over 50 people who met to present scientific work during a poster session.  Only parts of the poster session has been recorded. Around 6 hours of data have been recorded overall. The data was recorded from 2 top-down cameras that were each recording half of the scene simultaneously. The cameras are placed approximately 15 meters away from the participants of the event. The event consisted of a setting up period, coffee break, poster session, and taking down the posters at the end. The conversations that are formed were all natural and unscripted, motivated only by each individual’s real relationships with other people at the event. The data captures interactive behaviour between people who are closely acquainted (e.g. work colleagues or friends), as well as strangers.

It shows various different types of activities such as putting up posters, presenting posters, chatting, and getting tea and coffee.

 The distance and orientation of the camera helps to conserve the privacy of the participants while enabling the analysis of a crowded scene. All frontal or semi-frontal faces have been blurred.

 

Annotations of the data for F-formations (conversing groups), body position and orientation information for a subset of the data are available by contacting Hayley Hung directly.

 

Further information about the data can be found in this paper, which should also be cited on the use of both the data and any annotations that are derived from it.

 

H. Hung and B. Krose, "Detecting F-formations as Dominant Sets", in International Conference on Multimodal Interaction, Alicante, Spain, November 2011.

 

Format of the Data:

00006706.jpgThe video has been extracted into frames and put into several folders. There were two cameras (A, which shot the left part of the scene and B, which shot the right part). The data was recorded with tape and since the tapes needed to be changed half way, the scene might have shifted a bit.  For each camera, there are three consecutive parts. For example, CamA_part1_003 is the folder that contains image frames that follow from CamA_part1_002. Part 2 tends to be more crowded while parts 1 and 3 are more sparse.

 
This data was recorded by Hayley Hung with assistance from Bastien Crettol, Jaggan Varadarajan, Valerie

Devanthery, and Lu Zhang.

Document Actions