Discussion questions from CS 7390 lecture

Date: 2/28/96

Presented by: Greg Newton

Topic: 3D Software Visualization

• In Stasko's paper, he categorizes 3D visualizations into three categories (augmented 2d views, inherent 3d data sets, and adapted 2d views). Brown also has three categories: additional info about 2d objects, uniting multiple views, and capturing history of execution. Clearly, Stasko's 1st and 2nd categories (augmented 2d views & 3d data sets) don't overlap with any of Brown's categories. Does the other categories presented by Stasko completely overlap with the categories presented by Brown, or has Brown identified new uses for 3d visualizations? Also, does augmenting 2d views just add to the ``beauty'' of a picture as Brown says, or does it make the animation easier to understand?
• I believe that the argument ``3D is useful when there is not enough screen real estate'' is fundamentally flawed, because:
  • 3D views must still be projected to a 2D plane. If this projected view sufficiently conveys the necessary information, then any 2D view will do the same.
  • 3D requires higher resolution than 2D. TO understand the 3D shape of an object, many pixels are required. To make full use of lighting, large surfaces are required. In addition, obscured data further limits the information to pixel ratio.
  • 3D is useful to humans because we can often use direct manipulation of objects as well as our own viewpoint. Without a good navigation interface, 3D views are often less effective than 2D views. Note that today's VR is not the solution as we have to give up several factors of resolution.
  • Without stereo projection and proper lighting such as shadows, 3D views are often hard to understand--watch out for illusions.
  Unless the data is inherently 3D, 2 2D views are often better than one 3D view, and offer a better information per pixel ratio. In most instances, I'd strongly vote against the use of ``2D adapted views.''
• Stasko & Wehrli say that the traditional 3D animation tools are not well suited for animating computation. This may have been true in the past, but I think tools like Alias have become sophisticated enough and flexible enough to handle the task. But the point about the learning curve for these packages is quite valid. Under what circumstances would it make sense to try to use a production-quality animation package for animating algorithms?
• Polka-3D offers convenient 3D objects and viewing for visualization. However, the role of dealing with the 3D programming still has to be dealt with by the programmer, i.e. as Polka moves to 3D the programmer must move to 3D too, an uphill learning curve. How can a system support this transition (for the programmer)?
• The Stasko and Wehrli paper discussed the use of VR in visualizing 3D animations. So far there's no conclusive evidence that algorithm animations help people learn the algorithm and there's no conclusive evidence that VR helps people learn. Is there a real belief that combining the two will be beneficial?
• What kind of viewpoint control does Polka-3D provide, besides the virtual environment approach?

• Brown suggests that 3d animations can help to unite multiple views. He then gives examples of 2 algorithms (heapsort and kd trees) where he has done this. Consider the heapsort algorithm. I think it is important to realize that (in his solution at least) there is no angle from which you can view the 3d tree which will show both views at once that were initially separate and 2d. So you are losing some information in the 3d visualization (gaining some info, too of course but ...).
• This paper is very similar to Stasko's, except the figures are even more confusing. What 3D techniques are useful in alleviating this problem? I suggest that gridding and the use of shadows might enhance the depth cues.
• Whilst viewing the heap sort in the video, it wasn't clear to me that the animation was easy to understand. 3-D seemed to be overkill for this application.
• In Brown's paper, the claim is made: that 3D views allow 2D views to be combined to create a more intuitive combined 3D view. However, combining these views actually confuses the information (in many cases). Could multiple 2D views (in addition to the 3D) be added to supplement the 3D as well? This would be similar to multiple (redundant) views in a CAD system.
• Brown says 3D visualization can allow multiple views to be combined into one. What kinds of views cannot be combined (possibly due to lack of physical layout commonalities) and what can be done to overcome this problem, if anything?
• In Najork & Brown paper, they limit themselves to a sort of extended 2D views and don't want to represent intrinsically 3D objects. Why so? Software limitation? Is it hard to find appropriate application?

• Both papers dismiss the ``inherently'' or ``intrinsically'' 3D applications as not being as interesting, but I disagree. What ideas can be carried over from these intrinsically 3D apps to help in putting together an ``adapted 2D'' view?
• What steps can be taken to avoid visual ``clutter'' that can arise from complex 3D views? Do the advantages of 3D justify the extra effort to make sense of what you are looking at?
• It seems that the Brown paper does not address Stasko's categories of inherent 3D application domain views which I feel is of prime importance.
• Both the Brown and the Stasko papers divide 3D visualization into 3 areas--although these areas are different for the 2 papers. What are the similarities/differences and are there any other areas neither of them mention?
• Does animation become more important when you use 3D? It seems that simply flashing or ``warping'' of objects from one place to another will be misleading since it's not as easy to visualize the 3D spatial relations.
• Since 3D visualizations are mapped onto a plane (the screen) it could be argued that the only way to truly take advantage of 3D is if the view is rotated during the animation. How would this limit the examples given in these two papers? Limit 3D visualization in general?
• Neither of the 2 papers discussed the real uses for 3-D animations of algorithms--in terms of a learning tool vs. furthering understanding. Take an algorithm in 2D that has been ``augmented'' to use 3-D and show additional information in this third dimension as an example. Does an animation that provides more information necessarily help a person learn an algorithm or is there great opportunity to be overwhelmed by information than with 2-D?
• 3D algorithm animation utilizes the 3rd dimension to visualize an extra variable in the algorithm, for example state information and history. But the display device is inherently 2D, so there are certain view points that are more meaningful than the others. How can we help to identify these more meaningful view points?
• Where do we draw the line between using 3D to add useful information and just making it look pretty? As with color, some use can make a viz easier to follow, but gratuitous use can make it even harder to read--hence the ``angry fruit salad'' syndrome.
• There is not much critics about 3D. Does it really give an edge, improve understanding as it is claimed? How about some user-testing?