ChildLens: An Egocentric Video Dataset for Activity Analysis in Children

We present ChildLens, an egocentric video and audio dataset capturing naturalistic everyday experiences in children aged 3–5 years and including detailed activity labels. A total of 109 hours of experiences were recorded from 62 children in their home environment using a 140° wide-lens camera equipped with a microphone embedded in a child-friendly vest. Annotations include five location classes and 14 activity classes, covering audio-only, video-only, and multimodal activities. Captured through a vest equipped with an embedded camera, ChildLens provides a rich resource for analyzing children’s daily interactions and behaviors. We provide an overview of the dataset, the collection process, and the labeling strategy. Additionally, we present benchmark performance of two state-of-the-art models on the dataset: the Boundary-Matching Network for temporal activity localization and the Voice-Type Classifier for detecting and classifying speech in audio. Finally, we analyze the dataset's specifications and their influence on model performance. The ChildLens dataset will be freely available for research purposes. It provides rich data to advance computer vision and audio analysis techniques and thereby removes a critical obstacle to studying the context of child development.