The Oomplet Dataset Toolkit: A flexible and extensible system for large-scale, multi-category image generation

Understanding the dynamics of perceptual learning in humans, non-human animals, and artificial agents requires large stimulus sets with flexible features that can be used to discriminate across categorical groups. Here we introduce the Oomplet Dataset Toolkit (ODT), an open-source, publicly available toolbox for generating up to 9.1 million unique visual stimuli that are assembled across ten different feature dimensions. The resulting stimuli consist of cartoon-like humanoid characters -- ''Oomplets'' -- that are meant to be engaging, pleasant to look at, and can be appropriately used in research on a variety of populations, including children. Across several behavioral experiments, we show how eight of the ten possible dimensions that define an individual Ooomplet can be used by adults as effective classification boundaries for simple perceptual discrimination. The ODT thus provides a flexible and customizable way for generating very large and novel stimulus sets in order to study perceptual learning in both biological and artificial systems.