Fig. 1. HALS: human-augmenting AI-based labeling system.

As a human annotator labels data, an active learning algorithm shuttles the annotator around the image by identifying the next best visual features to annotate. Simultaneously, other AIs make labeling suggestions designed to significantly accelerate annotation speed (not pictured—see Fig. 3. System Architecture). Together, they give annotators the ability to train personalized AI models, enabling them to generate high-quality labeled data sets for otherwise intractably large images. See Data Availability for image details.