Family in the Wild (FIW): A Large-scale Kinship Recognition Database

We introduce a large-scale dataset for visual kin-based problems, the Family in the Wild (FIW) dataset. Motivated by the lack of a single, unified image dataset available for kinship tasks, our goal is to provide a dataset that captivates the interest of the research community, i.e., large enough to support multiple tasks for evaluation. For this, we collected and labelled the largest set of family images to date, with only a small team and an efficient labelling tool that was designed to optimize the process of marking complex hierarchical relationships, attributes, and local label information in family photos. We experimentally compare our dataset the existing kinship image datasets, and demonstrate the practical value of the newly collected FIW dataset. We also demonstrate that using a pre-trained convolutional neural network (CNN) as an off-the-shelf feature extractor as performing better than traditional feature types used for kinship based tasks in the visual domain. We also measure human performance and show their performance does not match up to that of machine vision algorithms.