A fundamental difference between rodent and primate object vision

Abstract

Studies of the rodent visual system have exploded in recent years, revealing tuning to elementary image features similar to primates. How are these features further transformed? In the primate, researchers have argued that a key transformation is ?figure-ground segmentation.? Behaviorally, humans can delineate figures from backgrounds using only kinetic borders independent of low-level texture cues. Neurally, long-latency enhancement of responses in early visual areas to regions within figure compared to background has been reported. Do mice also show behavioral and neural signatures of figure-ground segmentation? We trained mice on a figure-ground segmentation task where figures were defined by gratings and naturalistic textures, moving counterphase to the background. Unlike primates, mice were incapable of figure-ground segmentation using kinetic borders alone. While mice could report the location of gratings, they were at chance for figures defined by naturalistic textures. Remarkably, mice could learn to localize naturalistic texture-defined figures after many weeks of training, but when tested on new textures, were unable to generalize. This suggests a strategy of memorizing a lookup table of noise patterns, a cognitively impressive feat. We also recorded visual responses to the same stimuli in V1, RL, and LM using both 2-photon imaging and electrophysiology. Neural responses were consistent with the behavior, revealing robust position decoding from neural populations for gratings but not naturalistic textures. Lastly, we simulated neural responses using a model of V1 incorporating orientation-dependent suppression, and a deep network, VGG16. While the former model failed to reproduce the neural results, the latter model was successful. Interestingly, the best model for mouse V1 neural data came from mid to late layers. Taken together, these findings reveal a fundamental difference between primate and mouse mechanisms for object segmentation, showing that orientation contrast is critical for object perception in mice, unlike in primates.