Section: Scientific Foundations
A failure of standard modeling techniques?
Surprisingly, in our digital age, conceptual design of static shapes, motion and stories is almost never done on computers. Designers prefer to use traditional media even when a digital model is eventually created for setups such as industrial prototyping, and even when the elements to be designed are aimed at remaining purely virtual, such as in 3D films or games. In his keynote talk at SIGGRAPH Asia 2008, Rob Cook, vice president of technology at Pixar Animation Studios, stressed that even trained computer artists tend to avoid the use of 3D computerized tools whenever possible. They use first pen and paper, and then clay to design shapes; paper to script motion; and hand-sketched storyboards to structure narrative content and synchronise it with speech and music. Even lighting and dramatic styles are designed using 2D painting tools. The use of 3D graphics is avoided as much as possible at all of these stages, as if one could only reproduce already designed material with 3D modelling software, but not create directly with it. This disconnect can be thought of as the number one failure of digital 3D modelling methodologies. As Cook stressed: “The new grand challenge in Computer Graphics is to make tools as transparent to the artists as special effects were made transparent to the general public” (Cook 2008). The failure does not only affect computer artists but many users, from engineers and scientists willing to validate their ideas on virtual prototypes, to media, educators and the general public looking for simple tools to quickly personalize their favourite virtual environment.
Analyzing the reasons for this failure we observe that 3D modeling methodologies did not evolve much in the last 20 years. Standard software, such as Maya and 3dsMax, provide sophisticated interfaces to fully control all degrees of freedom and bind together an increasing number of shape and motion models. Mastering this software requires years of training to become skilled. Users have to choose the best suited representation for each individual element they need to create, and fully design a shape before being able to define its motion. In many cases, neither descriptive models, which lack high level constraints and leave the quality of results in user’s hands, nor procedural ones, where realistic simulation comes at the price of control, are really convenient. A good example is modelling of garments for virtual characters. The designer may either sculpt the garment surface at rest, which provides direct control on the folds but requires lots of skill due to the lack of constraints (such as enforcing a cloth surface to be developable onto a plane), or they can tune the parameters of a physically-based model simulating cloth under gravity, which behaves as a black box and may never achieve the expected result. No mechanism is provided to roughly draft a shape, and help the user progressively improve and refine it.
Capture and reconstruction of real-world objects, using either 3D scanners or image-based methods, provides an appealing alternative for quickly creating 3D models and attracted a lot of attention from both Computer Graphics and Computer Vision research communities the last few years. Similarly, techniques for capture and reuse of real motion, enabling an easy generation of believable animation content, were widely investigated. These efforts are much welcome, since being able to embed existing objects and motion in virtual environments is extremely useful. However, it is not sufficient. One cannot scan every blade of grass, or even every expressive motion, to create a convincing virtual world. What if the content to be modelled does not exist yet, or will never exist? One of the key motivations for using digital modelling in the first place is as a tool for bringing to life new, imaginary content.