Lead author Ramanujan Srinath showed that 3D shape is explicitly represented in visual area V4, just two stages beyond primary visual cortex. 2-photon microscopy revealed that 2D and 3D shape processing are segregated into patterns of interleaved modules across the surface of area V4. Remarkably, similar representations of 2D and 3D shape were found in AlexNet, a standard deep network visual system. These results suggest that brain evolution and deep network learning have independently discovered that early transformation of 2D eye/camera images into representations of 3D reality is an optimal strategy for object vision.
See Dispatch describing these results in the journal: