Computer Graphics Qualifying Exam

September 24, 2001

This exam consists of six questions. You are required to answer five questions in the time allotted. Answer each question in a separate blue book.

  1. In 1988, Forsey and Bartels introduced Hierarchical B-Splines. This technique was extremely important because it added a new shape representation with a key feature. More recently, subdivision surfaces have emerged as a useful and popular surface representation, because they too offer advantages over previous methods.

    1. What is the key feature that Hierarchical B-Splines provided (over shape representations that came before them)? Explain why is this so significant, using an example to demonstrate your argument.
    2. Describe two difficulties with spline based surface representations that are overcome with subdivision surfaces. Explain why these tasks are difficult with splines.

  2. A graduate student is planning to write a renderer that will be very photo-realistic. Her goal is to produce images that depict as many of the lighting effects seen in the real world as possible, using only the current state of the art in image synthesis algorithms.

    Her advisor (who obviously knows little about graphics) insists that she write her renderer without using any random number generator, as he believes this will mean her results are more reproducible. This precludes her use of several modeling and rendering algorithms.

    Name some algorithms that she will not be able to use in her renderer. Describe the effect that absence of each algorithm will have on the resulting system's performance (in terms of image quality, computation time, effects that can be produced, ...)

  3. In his 1997 paper, Peter Litwinowicz discusses many issues in creating "painterly" video, and suggests computer vision algorithms that allow for the automation of this task.

    In a sequence of papers beginning with Winkenbach and Salesin's 1994 paper, the simulation of a pen and ink style has proven to be a popular "artistic style."

    Describe issues that would arise in doing "animated pen-and-ink", and suggest how methods that appear in the stylized rendering literature, such as those proposed by Litwinowicz, might be applied.

  4. The well-known REYES rendering architecture (which is the "standard" implementation of Renderman) was introduced by Pixar scientists as an efficient and effective way to produce very high-quality images. REYES introduced a new rendering algorithm called mico-polygonalization in which geometric primitives were broken into small (sub-pixel) sized polygons before being considered for shading. Over the years, this style of "micropolygon-based" renderer has come in and out of favor as the best choice for doing high-quality renderings.

    Current trends in computing hardware might make a micropolygon renderer a good choice. Discuss why this is true. Hint: consider how a micropolygon-based renderer might be implemented on modern graphics hardware.

  5. Physical simulation is a good mechanism for creating some kinds of motion in computer graphics, and a terrible one for other settings.

    Describe four real-world physical effects for which (given the current state of the art) physical simulation is an appropriate method for creating motion, and situations in which each might be used. Give examples of at least two situations where simulation would be a poor tool for creating motion, and the reasons why.

  6. The Quicktime VR system, the Light-Field and Lumigraph systems, and Debevec's Facade system represent three different approaches to image-based modeling and rendering. For each approach, describe a situation to which it is well suited. Also explain why the other approaches are not suited to each situation.
    Your answer should consist of a description of three scenes, and for each situation a description of why each system would or would not be appropriate.