Shape conceptualization vs shape recognition
Although humans count with biological 3D sensors like stereographic view and ocular focus, much of the image interpretation is achieved through conceptualization
Indeed when a human watches TV or plays a computer simulation game or remotely pilots a vehicle through a screen, it is the conceptual vision the only mechanism in charge of building the 3D scenes on his mind because no other 3D information is actually available.
The sketch interpreter of the LAI4D´s design assistant is the face of the conceptualizer program JAIC. The goal of this program is to “imagine” the conceptual geometry represented by an imprecise raster image in a probabilistic way regardless of whether it is a recognizable shape or not.
The sample image shows a set of irregular strokes at the left side, and it is what it is, a sketch with irregular strokes. Despite their imprecision, a person could also say that the strokes represent a box, like the one at the right side. The necessary process to reach to the conclusion that those irregular strokes are representing a perspective view of a box is a conceptualization process. While a box is a well-known shape, the conceptualization can also be extended to the comprehension of non-recognizable and never-seen-before geometries when such geometries have some kind of sense.
JAIC exploits artificial intelligence in a very unconventional way due to it does not use recognition but conceptualization. This video demonstration of the LAI4D’s designer 2.0 shows how sketch conceptualization can work with general shapes:
A shape recognition process attempts to find a match for the analyzed image (or other input) within a shape descriptions database returning as result the found record if any. However, a shape conceptualization process attempts to “imagine” the shape represented by the analyzed image regardless of whether it is a recognizable shape or an abstract one returning the imagined shape’s geometry as result.
This is a good example of shape recognition applied to design:
The introduced application is able to recognize a predefined set of sketches each one associated to the corresponding 3D shape.
This other one is a more direct example of a sketch-based shape recognition system aimed to find a 3D geometry within a database from a sketch:
While image recognition based on convolutional neural networks can be extremely useful, it is intrinsically limited. As a detail JAIC only uses recognition for the OCR function. Convolutional neural networks are very good for finding similarities between an input and a database of patterns, like in the OCR, but this approach is not enough to solve the problem of conceptualizing generic images. In the other hand, JAIC implements image conceptualization that has no specific limitations. Unfortunately this innovative and promising technology is at a early stage of development and far away from its full potential.
This video shows more illustrative examples of the LAI4D’s 3D sketch interpreter capabilities:
More information on the LAI4D’s official web site.
