Abstract
Color is a prime example of categorical perception, yet it is unclear why and how color categories emerge. While prelinguistic infants and animals treat color categorically, several recent modeling endeavors have successfully utilized communicative concepts to predict color categories. Rather than modeling categories directly, we investigate the potential emergence of color categories as a result of acquiring visual skills. Specifically, whether color is represented categorically in a convolutional neural network (CNN) trained to recognize objects in natural images. Systematically training new output layers to the CNN for a color classification task, we find clear borders between new (non-training) colors that are largely invariant to the training colors. Using an evolutionary algorithm that relies on the principle of categorical perception we verify these border locations. These results provide strong evidence that color categorization emerges as a function of basic visual skills and provide a new basis for uncovering how they emerge.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Figure 2D revised and text in Invariant Border experiment adapted regarding circular correlation analysis; Different ResNets have been tested and included in supplemental document; Simulations of Invariant Border Experiment have been visualized and included in the supplemental document; General text improvements
