The SPOT Grid: Naturalistic images with predefined settings, patterns, objects, and textures

In the natural world, higher-level visual features (e.g., scenes and objects) systematically covary with specific lower-level constituent features (e.g., textures and patterns). This natural covariation makes it difficult to isolate aspects of behavior or brain activity that are specific to a single level of the feature hierarchy. For example, a particular neural or behavioral response may reflect recognition of a "jungle" scene, or it may reflect sensitivity to the lower-level features - leaf textures, dense vertical lines, etc. - that constitute jungle scenes. To address this challenge, we developed a novel approach for partially decoupling higher- and lower-level features in naturalistic images. We employed a Stable Diffusion image generation framework to create a naturalistic image set in which setting-object pairings and texture-pattern pairings are factorially combined, enabling systematic manipulation of higher-level and lower-level image content. To validate our approach, we used state-of-the-art image-processing models to confirm that higher-level features are represented similarly regardless of their lower-level constituents and, conversely, that lower-level features are represented similarly regardless of the higher-level objects/settings they constitute. We then confirmed that human participants can readily and rapidly categorize these features in our generated images. We present the image generation framework and introduce the SPOT (Settings, Patterns, Objects, Textures) Grid, a publicly available stimulus set accompanied by image-specific metrics (upon request: forms.gle/X2HpzRYUGigFS1S27). This resource provides new opportunities for investigating hierarchical visual representations, feature binding, and abstraction across perception, behavior, and the brain.