# Unraveling the complexity of rat object vision requires a full convolutional network and beyond

## TL;DR

Rats show advanced visual object recognition abilities that require a full CNN to match, especially when objects are partially hidden or simplified.

## Contribution

The study reveals that rat object vision is more sophisticated than previously thought, requiring CNNs to model their tolerance to complex visual transformations.

## Key findings

- Rats' tolerance to translation, scaling, and rotation is matched by half of a CNN.
- Full CNN is needed to match rat performance with occluded or outlined objects.
- Rats use more view-invariant discrimination strategies than CNNs.

## Abstract

Despite their prominence as model systems of visual functions, it remains unclear whether rodents are capable of truly advanced processing of visual information. Here, we used a convolutional neural network (CNN) to measure the computational complexity required to account for rat object vision. We found that rat ability to discriminate objects despite scaling, translation, and rotation was well accounted for by the CNN mid-level layers. However, the tolerance displayed by rats to more severe image manipulations (occlusion and reduction of objects to outlines) was achieved by the network only in the final layers. Moreover, rats deployed perceptual strategies that were more invariant than those of the CNN, as they more consistently relied on the same set of diagnostic features across transformations. These results reveal an unexpected level of sophistication of rat object vision, while reinforcing the intuition that CNNs learn solutions that only marginally match those of biological visual systems.

•Rat object vision is modeled using a deep convolutional neural network (CNN)•Half the CNN is needed to model rat tolerance to translation, scaling, and rotation•The whole CNN is required to match rat performance with occluded and outlined objects•The view-invariance of the rat discrimination strategy is unmatched by the CNN

Rat object vision is modeled using a deep convolutional neural network (CNN)

Half the CNN is needed to model rat tolerance to translation, scaling, and rotation

The whole CNN is required to match rat performance with occluded and outlined objects

The view-invariance of the rat discrimination strategy is unmatched by the CNN

Rodents have become very popular model systems to study the neuronal mechanisms underlying visual perception, yet it remains unclear how advanced their visual system really is. Here, we use a deep neural network to understand how computationally difficult it is for rats to succeed in different object-recognition tasks. We found that at least half of the network is required to match rat perception when objects undergo spatial transformations, such as translation and scaling. The whole network is needed instead to equate rat ability to recognize the target objects despite partial occlusion and reduction to outlines. In addition, rats process visual objects by using discrimination strategies that are more consistent across view changes than those of the network. These findings reassert the sophistication of rat vision while suggesting interesting adjustments to the architecture and visual diet of deep nets to bring their robustness and generalization power closer to their biological counterparts.

Muratore et al. modeled rat object vision using a deep convolutional neural network (CNN). They found that processing up to half of the CNN is needed to account for rat tolerance to translation, scaling, and rotation of target objects, while the whole network is required in the case of partial occlusion and reduction to outlines. Instead, the complexity and view-invariance of rat discrimination strategy remains unmatched by the CNN. These findings indicate that rats process visual objects by using advanced neuronal machinery.