This dataset contains raw and pre-processed images created for the paper "Zebrafish identification with deep CNN and ViT architectures using a rolling training window" by Puchalla et al. The images in this dataset was used to train machine learning models for minimally invasive individual juvenile zebrafish identification, as the zebrafish matured over several weeks. Images were captured in situ using an acrylic aquarium insert, allowing the fish to remain free-swimming. The images consist of five zebrafish and were captured on five different days, spanning a total of 19 days.