“I sculpt shadow with light or sometimes light with shadow, but both function in essentially the same manner. I take objects and carve and place them in relation to a single light source. The complete artwork is therefore comprised of both the material (the solid objects) and the immaterial (the light or shadow). - Kumi Yamashita

When viewing Kumi Yamashita’s ingenious art, we confront our simple notion that solid predicts shadows. The artist in her work wants to question all our expectation, so she starts challenging the predictable relationship between solids and their shadows. From simple beginnings, she invites us to reassess the unpredictable relationship between what we expect to see and our actual perception.

Much of Kumi Yamashita’s artwork is constructed with everyday things such as alphabets and numbers (Figure 13), building blocks, thread, nails, credit card rubbings and light installations. As a result of the process these ordinary material objects are transformed into arresting yet elusive images. Separate pieces may connect together in a shadow, and a simple piece of paper can create an endless variety of profiles.

Kumi’s methods and materials go beyond the confines of traditional media, transforming one medium into something else.  With great attention to detail, Yamashita’s works are exhaustively complex and precise— yet they remain deeply human.

Figure 13, Kumi Yamashita, City View, 2003, H250, W500, D5cm. Aluminum numbers, single light source, shadow. Commissioned by Namba Parks Tower, Osaka Japan. Kumi Yamashita, 2012, http://www.kumiyamashita.com.

Figure 14, Kumi Yamashita, Conversation, 1999 H150, W40, D40cm Styrene, motor, single light source, shadow. Kumi Yamashita, 2012,

By constructing carefully measured sculptures realistic shadows of people are formed when light is projected past them. In the piece called Conversation (or Dialogue) (Figure 14), we see a motorized construction, which produces the illusion of a talking head.
Each sculpture is a masterpiece that leaves amazing shadows on the walls and each of us can find something new in these shadows, while this sculpture is nothing more than pieces of materials joined together to make the breathtaking chaos. As usual, the artists use ordinary stuff to create the shadow art sculptures and the projector helps the viewers to turn on their imagination and see the projected shadows on the wall.
Yamashita's sculptures illustrate the problem—any number of objects could produce the same retinal projection, just as the same shadow could be produced by many different sculptures. What's really interesting about vision is that the shortcuts and heuristics our visual system relies on to make this inference get it right so often. In The Ecological Approach to Visual Perception, Gibson speaks rather disparagingly of what he calls "aperture vision"

“The eye is easily deceived, and our faith in the reality of what we see is therefore precarious. For two millenniums we have been told so. The purveyors of this doctrine disregard certain facts. The deception is possible only for a single eye at a fixed point of observation with a constricted field of view, for what I called aperture vision. This not genuine vision, not as conceived in this book. Only the eye, considered as a fixed camera can be deceived. The actual binocular visual system cannot.”

Gibson's approach to the study of perception emphasizes the way an active observer picks up information from the environment. The central postulates of Gibson's approach are that visual space is defined by information (such as texture gradients) contained on environmental surfaces, the crucial information for perception is information that remains invariant as an observer moves through the environment, and this invariant information is picked up directly, so that no intervening mental processes are necessary for visual perception.  

I tend to think that most objects are defined not by their static properties, but by how those properties remain constant (and vary) across changes in view. My problem with the idea that the visual system just "picks up" such invariants is that it has been remarkably hard to identify invariant properties that could define objects in the world in a tractable way. It's easy to show that motion disambiguates, but it's hard to show that something other than the retinal projection is the information source for vision. For that reason, I tend to think that the retinal image and changes to it are the starting point for vision.
The flaw in the traditional approach comes not from thinking about the 2D-3D inference problem, but from focusing solely on static images rather than addressing how dynamic changes in view and the accompanying changes to the retinal projection of the world help resolve the inference problem.