Skip to main content
. 2023 Mar 23;26(4):606–614. doi: 10.1038/s41593-023-01280-0

Extended Data Fig. 2. Predictive coding model trained with an alternative set of natural images.

Extended Data Fig. 2

We used natural images of animals and landscapes of the African savanna from van Hateren repository, used in51 (a) Top - RFs optimized at different levels of SNR. Bottom – center size (purple line) and relative surround strength (black line) plotted as a function of the SNR. (b) Top - RFs optimized at different levels of the vertical SNR asymmetry. Bottom – vertical surround asymmetry plotted as a function of the vertical SNR asymmetry of the model photoreceptor output (RGC input). (c) RFs predicted for different positions within the visual field. (d) Horizontal cross-sections of model RFs in (c).