CNS*2026 Halifax

Introduction
Visual textures, like blades of grass or bark on a tree, are pervasive in the natural world. These patterns, characterized by statistical regularities across spatial scales, help animals navigate the world and categorize their surroundings[1]. Textures are quite complex, yet can be readily synthesized and parameterized by computational models, hence they offer a useful entry point for studying visual processing at multiple levels: from the encoding of complex image statistics to the formation of invariant representations [2]. However, the circuit-level implementation of these computations in the brain remains poorly understood.


Methods
As part of the Openscope initiative, we present a new open dataset [3,4] consisting of simultaneous two-photon calcium imaging across four distinct regions of the mouse visual cortex and two imaging planes of mice engaged in a texture discrimination task. We investigate how different families of textures are processed before, during and after learning a texture discrimination task. We examine how population level representations of different classes of textures are encoded within the visual cortex.


Results

Results suggest that internal representations of textures emerge during learning (particularly in layer 5 across visual areas V1, LM, and AL), and mirror behavioral discriminability, with these encodings being high dimensional. Furthermore, we find that these representations can be updated, generalizing to a wide set of images as task complexity increases. Interestingly, family-specific internal representations appear to be task-dependent, as during passive viewing, neural responses are more selective to individual images than  to families.



Discussion
Together, these results suggest that texture category representations across visual cortical areas are not fixed, but are dynamically regulated by task engagement. This is consistent with a top-down attentional mechanism impacting the encoding of naturalistic stimuli, rather than being an innate property of the visual system. These findings highlight the importance of behavioral contexts in sculpting cortical population codes.


References

1.Li, A., & Zaidi, Q. (2000). Perception of three-dimensional shape from texture is based on patterns of oriented energy. Vision Research, 40(2), 217–242. https://doi.org/10.1016/S0042-6989(99)00169-8
2.Portilla, J., & Simoncelli, E. P. (2000). A parametric texture model based on joint statistics of complex wavelet coefficients. International Journal of Computer Vision, 40(1), 49–70. https://doi.org/10.1023/A:1026553619983
3.Ager, K., Akella, S., Bawany, A., Bennett, C., Dichter, B., Ghosh, S., . . . Williams, A. (2024). The OpenScope Databook (v1.2.0) [Software]. Zenodo. https://doi.org/10.5281/zenodo.12614664
4.DANDI:001461 [Dataset]. DANDI Archive. https://dandiarchive.org/dandiset/001461


Acknowledgement
This work was funded by the US National Institutes of Health (NIH) U24NS113646. The imaging dataset was obtained as part of the OpenScope program, which is operated by the Allen Institute / Neural Dynamics. We thank the OpenScope steering committee for their support, the Allen Institute founders, Paul G. And Jody Allen, and Karel Svoboda, for their vision, encouragement, and support.