Humans are able to parse the complex flow of visual information into meaningful objects that can be categorized, recognized and acted upon. Recent research has clarified the neural representation of these objects through innovations in experimental design, in particular the dissociation of multiple factors such as visual features and category membership, and in computational modeling, in particular deep convolutional neural networks (DNNs). With the help of this recent research I will argue in favor of a feature-based categorical code of objects in human visual cortex, a code designed to represent categories through their features. Here I will make use of research illustrating the similarities between information processing DNNs and in the human visual brain (e.g., Zeman et al., 2019, BioRxiv; Kubilius et al., 2016, PLOS Comput. Biol.), as well as studies focusing upon differences due to the reliance of visual cortex to rely upon category-specific visual features (e.g., Bracci et al., 2018, BioRxiv).
