How does the brain represent the structure of a visual scene (the relations among items, e.g., the cardinality) independent of scene contents (the objects in the scene, e.g., item identity)? Thompson et al. propose a dual-stream neural network model based on the parallel pathways of the primate visual system.