This repo includes code for learning features from natural images
The mechanisms underlying the development of area V1 have been extensively studied. V1 is unusual among visual cortical areas, however, in that the majority of V1 cells respond well to the same basic shape โ an elongated edge or grating-like feature with varying preferences for orientation, spatial frequency, and other parameters. Following from this, the core of most V1 models consists of a set of Gabor filters with varying orientation and scale parameters densely tiling the visual field. In contrast, mid-level visual areas such as V2 and V4 contain neurons selective for a multitude of diverse shape and/or texture features, but little is known as to what drives the development of such diverse RF properties in these areas. Given that V1 is a major source of input to mid-level visual areas, we hypothesized that natural images that strongly activate local populations of V1 cells could provide the rich input statistics needed to drive the diversification process in extrastriate areas.