The Sesame Street Archive: a labeled image repository of educational children’s television, 1969-2018

Karol Sadkowski, Siyuan Guo, Chen Yu, Sophia Vinci-Booher

Keywords: Sesame Street Archive (SSA), educational media, children’s television, object detection, face detection, image annotation, domain adaptation, child-centered scenes

TL;DR: The Sesame Street Archive (SSA) is a labeled dataset of over 35,000 images spanning 50 years of Sesame Street public programming, revealing key challenges in domain adaptation and model performance on children’s educational media.

Abstract: The Sesame Street Archive (SSA) is the first image repository based on educational children’s television, comprising over 35,000 primary-source film frames extracted from 4,397 Sesame Street episodes broadcast from 1969 to 2018. Using Sesame Street, perhaps the most enduring and internationally adapted educational television series in the world, the SSA provides the computer vision community with curated scenes of sociocultural and educational importance televised to children throughout six historical decades. SSA film frames contain highly variable object instances, including diverse characters, languages, numeric patterns, and built environments, situated in both real-world and stylized settings. This first-look report details the creation of the SSA with a training dataset of images to showcase its breadth and potential. The SSA will be made available for research only. The training dataset discussed in this report is temporarily accessible via https://github.com/muppetAnon/ssa-cvpr2025.

Submission Number: 8