People in Social Context (PISC) Dataset

The People in Social Context (PISC) dataset is a new dataset that focuses on social relationships. It consists of 22,670 images of 9 types of social relationships. We provide annotation of the bounding boxes of all people, as well as the social relationship between all pairs of people in the images. In addition, we provide occupation annotation. For more details on the collection process and statistics of the dataset, please see our paper.

• annotation_image_info.json contains bounding box annotation and information about image source, image size, image id.
• domain.json contains annotation of the 3 types of coarse relationships: {intimate, not intimate, no relation}.
• relationship.json contains annotation of the 6 types of fine relationships: {friends, family, couple, professional, commercial, no relation}.
• occupation.json contains annotation of the occupation.
• domain_split and relation_split contain train/val/test split.
• after downloading all image parts (images-*), extract using: cat images-* | tar zx

The dataset can be applied, but not limited to the following research areas:

• social relationship study
• people detection
• occupation recognition

Please cite the following paper if you use the PISC dataset in your work (papers, articles, reports, books, software, etc):

• J. Li, Y. Wong, Q.Zhao, M. Kankanhalli
  Dual-Glance Model for Deciphering Social Relationships
  ICCV, 2017.
  http://doi.org/