Capsule Networks but not Classic CNNs Explain Global Visual Processing

Classically, visual processing is described as a cascade of local feedforward computations. Feedforward Convolutional Neural Networks (ffCNNs) have shown how powerful such models can be. However, using visual crowding as a well-controlled challenge, we previously showed that no classic model of vision, including ffCNNs, can explain human global shape processing (1). Here, we show that Capsule Neural Networks (CapsNets; 2), combining ffCNNs with recurrent grouping and segmentation, solve this challenge. We also show that ffCNNs and standard recurrent CNNs do not, suggesting that the grouping and segmentation capabilities of CapsNets are crucial. Furthermore, we provide psychophysical evidence that grouping and segmentation are implemented recurrently in humans, and show that CapsNets reproduce these results well. We discuss why recurrence seems needed to implement grouping and segmentation efficiently. Together, we provide mutually reinforcing psychophysical and computational evidence that a recurrent grouping and segmentation process is essential to understand the visual system and create better models that harness global shape computations. Author SummaryFeedforward Convolutional Neural Networks (ffCNNs) have revolutionized computer vision and are deeply transforming neuroscience. However, ffCNNs only roughly mimic human vision. There is a rapidly expanding body of literature investigating differences between humans and ffCNNs. Several findings suggest that, unlike humans, ffCNNs rely mostly on local visual features. Furthermore, ffCNNs lack recurrent connections, which abound in the brain. Here, we use visual crowding, a well-known psychophysical phenomenon, to investigate recurrent computations in global shape processing. Previously, we showed that no model based on the classic feedforward framework of vision can explain global effects in crowding. Here, we show that Capsule Neural Networks (CapsNets), combining ffCNNs with recurrent grouping and segmentation, solve this challenge. ffCNNs and recurrent CNNs with lateral and top-down recurrent connections do not, suggesting that grouping and segmentation are crucial for human-like global computations. Based on these results, we hypothesize that one computational function of recurrence is to efficiently implement grouping and segmentation. We provide psychophysical evidence that, indeed, grouping and segmentation is based on time consuming recurrent processes in the human brain. CapsNets reproduce these results too. Together, we provide mutually reinforcing computational and psychophysical evidence that a recurrent grouping and segmentation process is essential to understand the visual system and create better models that harness global shape computations.