Humans easily recognize objects and perceive their attributes such as shape, color,
and material. To achieve these high-level functions, it has generally been assumed
that the visual system must reconstruct three dimensional surfaces from the retinal
image via deep neural computations. However, recent advances in computer vision
and psychophysics lead to an alternative scheme that the human visual system utilizes
low-level image features and their statistics directly to recognize 3D objects, scenes,
materials, and even arts. This heuristics-based vision requires shallow computation,
and is indeed suitable for quick and efficient comprehension of objects and stuffs in the
real world. However, such a shortcut is insufficient to explain our ability of ’looking
through the truth’ via careful scrutiny. The present paper reviews recent findings and
debates concerned with this paradigm shift.
