Dataset for Communication Context Identification

<b>Feature and label dataset from 68.9 hours of egocentric video of multi-conversation scenarios for Communication Context Identification (CCI): </b><b><i>Given an egocentric video stream with detected face bounding boxes, determine whether each individual is part of the camera wearer’s conversation group.</i></b>Dataset descriptionThe dataset was recorded in 6 recording sessions with 6-10 participants per session seated around a table. Overall, 48 individuals participated (22F, 1X, 25M; age mean = 25.5, std = 4 years; 3 native English speakers, 45 proficient second language speakers; no visual / auditory / speech / psychiatric conditions / impairments that may affect conversations).Egocentric video (25fps, 1080p) was recorded from each individual for 20 ~5-minute conversations with varying conversation partners, conversation group sizes, seating arrangements and background noise.The present dataset represents features for 68.9 hours of egocentric video. This does not include missing data from hardware issues or excluded segments when participants did not adhere to the specified groups.The dataset is designed to facilitate replication of the results published in the paper. See the code repository https://github.com/dorszewski/cci for a usage example.For further details see the paper and the metadata description (in the readme.md).CitationDorszewski, Tobias, and Hjortkjær Jens. "Seeing Conversations: Communication Context Identification in Egocentric Video". <i>Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition</i>. 2026Ethics approvalAll participants were financially compensated for their time and gave written informed consent. Data collection was approved by the Science-Ethics Committee for the Capital Region of Denmark, reference H-16036391.