Arqspin takes an image-based approach to 3D modeling. The appearance of a 3D object is represented as a set of images recorded from viewpoints spaced uniformly around a circle centered on the turntable’s axis of rotation. Spin credit: Mr. Kevin Hurd.
What is “3D”?
Some of the press we’ve received (including some of our own marketing materials) use the term “3D” to describe the datasets that Arqspin produces. We thought this would be a good forum to explain the underlying theory behind how our technology works and give you our rationale for using this term. We agree that the term “3D” can be misleading given its common interpretation, but we’ll let you decide if what we’re doing is “really 3D” after considering some history and theory.
Image-Based vs. Explicit 3D Representations
Conventional methods for representing the appearance of a 3D object focus on mapping the shape of the object’s surface in the form of a triangle or quadrilateral mesh along with some mathematical model of the way the object interacts with incident light at its surface (scattering and absorption profiles). This includes the type of 3D models produced with standard computer aided design (CAD) software and the output of conventional 3D scanners (most often with the help of a laser). This is what most people associate with the term “3D.”
Arqspin takes a different approach based on the theory of Image-Based Modeling and Rendering (IBMR). An “image-based approach to 3D modeling” captures an object’s shape and material properties implicitly, in the form of resampled and processed images usually recorded with a camera. Our “spins” are essentially a collection of images of an object captured at viewpoints spaced uniformly around a circle centered on the turntable’s axis of rotation (see figure above). A major challenge with an image-based approach is capturing this set of images in the first place. We think we’ve come up with a pretty clever way of simplifying this key task.
A Brief Historical Perspective
The history of IBMR dates back to the commercial product QuickTime Virtual Reality (a.k.a, “QuickTime VR” or “QTVR”). QTVR is a file format developed by Apple in 1994, meant to support a QuickTime plugin for viewing outward facing 360-degree panoramas and inward facing 360-degree visualizations much like our spins. The task of capturing these datasets was left open and the QTVR format never gained real traction, as evidenced by the current lack of available viewers and native browser support. Apple’s QTVR system was described in two research papers published at the legendary ACM SIGGRAPH conference, the leading venue for academic work in computer graphics:
- Chen, S.E., Williams, L., “View Interpolation for Image Synthesis,” Proceedings of ACM SIGGRAPH 1993.
- Chen, S.E., “QuickTime VR — An Image-Based Approach to Virtual Environment Navigation,” Proceedings of ACM SIGGRAPH 1995.
These early ideas were formalized and expanded in two other research papers that followed in 1996:
- Levoy, M., Hanrahan, P., “Light Field Rendering,” Proceedings of ACM SIGGRAPH 1996.
- Gortler, S.J., Grzeszczuk, R., Szeliski, R., Cohen, M., “The Lumigraph.” Proceedings of ACM SIGGRAPH 1996.
The greatest benefit of an image-based approach is that it is fully general and can be used to represent any 3D scene. In contrast, most explicit approaches require the presence of a single well defined opaque surface (a “2-manifold” to be precise) and therefore cannot represent translucent objects (wax, skin, marble, etc.), semi-transparent objects (glass, certain plastics, etc.), and surfaces with fine geometric detail like hair or fuzz.
One basic drawback of an image-based approach is that it does not support the ability to display the object from arbitrary camera viewpoints. It is not possible, for example, to change the camera’s angle of incidence to better inspect the top or bottom of a spin. Doing so would require some type of “view interpolation” mechanism. Additionally, image-based approaches do not allow editing the surface geometry and material properties, nor can they be sent to a 3D printer for fabrication. These operations require an explicit representation of the object’s surface and material appearance.
These boundaries are continuously shifting as the field of computer graphics advances and new research results are published. Indeed, there is a growing list of academic papers that describe intermediate or hybrid representations. The goal is to maintain the benefits of an image-based model while allowing certain operations that had previously required an explicit mapping of the object surface. We’re very excited about adding some new innovations to this space as well…
So, is Arqspin 3D?
It depends on your perspective. An image-based representation is a widely recognized representation of a 3D object, although one that does not support the same range of operations that most explicit representations do (operations that most people admittedly expect from a “3D” representation). Perhaps a more important question is: “Does Arqspin support the application you have in mind?” If you want a dead simple and inexpensive way to let people interactively inspect a 3D object over the Internet then the answer is YES. In fact, it’s a much less expensive and faster way of achieving that goal than using CAD software or a 3D laser scanner. On the other hand, if you need something that allows printing a replica of some object using a 3D printer then the answer is NO (at least, for now!).