Conveying Structure
From CS294-10 Visualization Fa08
Lecture on Nov 5, 2008
Readings
- Smart Visibility in Visualization. Viola and Gröller (html)
- Using Deformation for Volumetric Browsing. McGuffin et al. (pdf)
- Interactive Image-Based Exploded View Diagrams. Li et al. (html)
Optional Readings
- Interactive Cutaway Illustrations of Complex 3D Models. Li et al. (html)
- Non-Invasive Interactive Visualization of Dynamic Architectural Environments. Niederauer et al. (html)
- Cutaways and ghosting: satisfying visibility constraints in dynamic 3D illustrations. Feiner and Seligmann (Springer)
- Volumetric Illustration: Designing 3D Models with Internal Textures. Owada et al. (html)
Seth Horrigan - Nov 03, 2008 06:14:15 pm
I am usually quite impressed by Maneesh's interactive exploded view diagram systems. This particular one is no exception. The fact that such a system works, and requires relatively little time to convert from a static 2D image into an interactive system is quite amazing. It is an excellent application of processing power. Nevertheless, the specific use case seems somewhat odd to me. The idea of creating a 2D exploded view of an object, then taking that view and converting it into a interactive exploded view seems like an odd path to take in search of an interactive exploded visualization. Other of his systems I have seen demonstrated started from a CAD models or the like, which seems to me to make sense as a course of construction.
This specific image-based system is definitely more innovative, but I would like to hear more on why specific one should start from already created images as required. The justification I saw was "using 2D images allows us to leverage the abundance of existing static exploded views commonly found in textbooks, repair manuals, and other educational material" and that 3D models are "in general much more difficult to acquire or build", but why not just start with 2D computer models? It seems like an arbitrarily unnecessary step to print out these models then read them back in. Or perhaps the intent is truly to use said 2D computer models directly (in which case, it might make sense just to render the pieces within the interactive visualization software directly), and I just misunderstood...
Also, the overview of "Smart Visibility" concluded, "Computer generated visualization still needs improvement to compete with hand crafted illustrations in terms of expressivity, harmony, or aesthetics." I cannot say that I actually saw an evidence of this within the paper itself. This sounds more like the authors' preconceived understanding of the field than an actual conclusion based on the current state of technology.
Lastly, the figures from the Deformations paper are more than slightly disturbing to my peace of mind.
Ketrina Yim - Nov 03, 2008 11:14:24 pm
I agree with Seth's comment regarding Volumetric Browsing. Accompanied by various images of opened 3D heads, the words "Spreader", "Leafer", and "Peeler" in the paper take on unsettling new meanings. While the visualization is rather effective, the deformations were not exactly pleasing to look at initially. Maybe I was overreacting, but my first impression of the gaping craniums was that they would probably give me nightmares. Even the Visible Human Project didn't elicit this kind of initial reaction. Perhaps the volumetric data deformations would be less distressing if the rendering was less photorealistic (not that blue is realistic for the interior of a human head).
Ketrina Yim - Nov 05, 2008 07:17:24 pm
The discussion of "good views" has made me realize how my view preferences have affected my artwork. I find that most of my work is in what we've described as the generic oblique view (except for when I draw clocks and houses, coincidentally), and as such it's the easiest view for me to draw. Frontal views are fairly easy to do as well (though I do need to work on my foreshortening), and the most difficult of all are top-down and bottom-up views (unless the object is a ship, seafaring or otherwise). I've reached the conclusion that the uneven nature of my skill is due to the way I prefer to look at things. I don't usually look at objects from directly above or below, and given the choice I do prefer the three-quarters view because I like being able to see the front and side at the same time. It seems so obvious now.
<shameless plug>If some evidence (other than my whiteboard antics) is necessary, you can take a look at my gallery.</shameless plug>
On an unrelated note, there seems to be a correlation between the illustration subject and the structure visualization method used. Technical illustrations tend to use cutaways and exploded views, while anatomical illustrations tend to be depicted with sections and ghosting. What might be the motivation behind this?
Nicholas Kong - Nov 08, 2008 05:31:26 am
I, too, very much enjoyed the exploded view paper. Referring to Seth's comment, I think that the purpose of the work was to facilitate the use of existing exploded view diagrams, especially those containing many parts. It would also seem that if you already were in possession of a 2D model, complete with all the depth/occlusion information, then their program would need very little user input to generate a fully satisfactory interactive exploded view diagram.
The volumetric displays were interesting, to say the least. It did seem that most of the interactions were designed with actual surgical operations in mind, so perhaps from a biologist's or doctor's perspective these visualizations are not as disturbing and more functional. The anatomist who used the system appeared to find it very useful, in any case.
Scott Murray - Nov 10, 2008 03:39:44 pm
The visual manipulation techniques presented in the paper "Using Deformations for Browsing Volumetric Data," while novel, attempt to resolve an issue that I think may be insoluble. The authors present the problem as the fact that, in 3D visualizations, objects "in front" get in the way of those "in back," so one can never truly view the entire data set at once.
First, although the deformation approach does make some progress in addressing this, it will always be the case that some visual elements occupying a 3D space will block the view of others. Making some elements of the visualization semi-transparent, as the authors mention, is one solution to this problem, but in order to get a clear view of something "in back," you will always have to remove the thing "in front," thereby tampering with the two elements' spatial relationship. The spatial relationship carries meaning within the visualization, so by shifting the elements in space, one distorts the meaning represented.
Second, this fact may not be a problem at all, at least for dynamic visualizations, where elements may be hidden, shown, made transparent or opaque, and moved away and back into place. By adding the dimension of time, it becomes safe to manipulate elements within the visualization without distorting the human user's understanding of the data shown.
James Hamlin - Nov 12, 2008 12:42:56 am
@Seth: Well, the idea, I think, is that we can take any 2D exploded view diagram we've already got and with a bit of manual work generate an interactive exploded view from it. The key thing to note is that the user might not have created the original exploded view herself. As you rightly ask, if you've already got a representation of the object, 2D or 3D, why not use that? Certainly you could, but for the vast corpus of exploded views that already exist, such models may be hard to acquire, may no longer exist, or may have never even existed. With this system we can take any of these and enrich it by turning it into one of these interactive exploded views.
As for the paper on visualizing volumetric data, I remain unconvinced of the utility of the "Sphere Expander" tool. There may well be a good use for it, but I don't think their example demonstrated one.
Maxwell Pretzlav - Nov 12, 2008 12:14:25 am
In an abstract sense, it seems to me that most of these approaches to viewing complex structure boil down to just a few different techniques. In my mind I divide most of these interactions into deformative and non-deformative modifications of a view, and into spacial and spectral modifications. Interactions either deform shapes or keep them in their original shape, or they either move/resize shapes, or change their opacity. The McGuffin paper seems to grasp this idea with its different widgets corresponding to different actions performed on an object; some deform parts of an object, while others move parts. The exploded views generated by Li et al is a sophisticated and complex system for preserving the shape and size of images and objects, so that they can be moved in specific useful ways.
It seems the choice between these different techniques is highly dependent on the subject of the visualization. Anatomical models lend themselves (although sometimes gruesomely) to deformation, as that has (excluding bone) similar analogues to reality. Additionally, often the goal may be to view a specific part in great detail. Mechanical diagrams, on the other hand, don't seem lend themselves to distortion—the context of a particular part of a mechanical system is too import to allow it to be distorted in pursuit of unobstructed visualization. Instead movement and opacity must be used. I wonder if there has been research into how well these different visualization techniques work depending on subject matter.
Michael So - Nov 12, 2008 12:50:33 am
I think it is interesting to point out the importance of animation in conveying structure. As demonstrated in the user feedback section of the Volumetric Browsing paper, animated transitions are "valuable". Without the animated transitions, the author(s) found out that people had trouble in understanding how the tools (such as the "Peeler" and "Leafer") worked. It is like an ellipsis, where the omission of the transition leaves the user to guess at how an image went from one state and ended on another state. No matter what the user guesses, what transition took place is still a mystery, so I think it is good to avoid such omissions. This revelation is also demonstrated in the interactive exploded-view diagrams. Directly manipulating the exploded-view diagram employs the use of animation. And this direct manipulation enhances understanding of the structure of the object such as how the individual parts of the object relate to one another. I guess animation is helpful in conveying structure because it reflects real-life motion.
David Poll - Nov 12, 2008 04:36:44 am
@Michael I'd like to wholeheartedly agree with you on the importance of animation in conveying structure. One of the recurring themes that I've gotten from the lectures throughout the semester has been that comparison, reference illustrations, and animations play an extremely important role in being able to comprehend visualizations. First, we saw how one visualization can give another enough context to be better understood. Then, in the context of the "spot the differences" experiment, we saw how important the transition between visualizations is. The corollary to that is the significance of animation, which combines a "series" of visualizations to produce a far more coherent message for a particular visualization. I don't want to overstate -- animation on its own isn't necessarily expressive, but as a supplement to a visualization, even tiny animations can make a world of difference.
Witton Chou - Nov 12, 2008 04:58:37 am
The interactive exploded view diagrams is great. If all the diy assembly furniture utilized this, I think people would be a lot less frustrated at the diagrams most come with. It also seems very easy to create an exploded view diagram through the demonstrations presented. One of the few downsides I see to companies using this is a loss in revenue from people purchasing professional assembly. In all seriousness, though, I think the interactive exploded view diagrams presented here are very useful and easy to use. Although this presents a 2/2.5D images, I think 3D models would still help more - though the cost/time efficiency would probably be significantly lower.
Matt Gedigian - Nov 12, 2008 11:23:05 pm
I enjoyed the Niederauer paper on non-invasive visualization. The exploded view visualization technique itself was described in more detail elsewhere, but in this case the interesting bit was the data-source. The direct manipulation of OpenGL commands using Chromium was clever -- I hadn't realized that tools for this existed.
The advantage is that this can be applied to any program that renders with OpenGL and requires only minimal user input (player height, number of floors, and the up vector). Unfortunately, using this low-level data means that we don't have all the useful semantic information that we might want to add to the new view (such as player information).
I'm wondering if the requirements for this have changed as game complexity and GPU performance have increased. Has the balance been maintained, so an 8-node cluster still suffices?
Chris - Nov 12, 2008 11:46:47 am
It's interesting to note the similarity between the two related research areas here. In "Using Deformation for Volumetric Browsing" and moreso in some related papers (see for example Interactive Cutaway Illustrations of Complex 3D Models), the authors are solving the problem of interactively manipulating medical data in a way that will produce views/visual representations that align with what the user will see when performing a real surgery. There is a related area of surgical simulation in physically based modeling in which, using synthetic data (3D models of anatomy) researchers try to create not only the visualization of interactive manipulation but also render the physical feedback that a surgeon would get through the use of haptic devices. A small article on this is at this NIH page.
Sarah Van Wart - Nov 12, 2008 11:52:00 am
[As others have mentioned] Yes, peeling, slicing, and spreading are grotesque in the context of human anatomy, but I doubt a surgeon would even bat an eye :). In fact, I found the professional anatomist's analysis of the various interactive volumetric data browsing techniques interesting (in the 'Browsing Volumetric Data' article). The anatomist found the peeling back of layers to be very helpful for exploration and contextualizating tissues and organs while doing a good job of maintaining the layers' spatial relationships. S/he even mentioned that it might not be necessary to display the peeled layers. As with many of the other visualization techniques mentioned in this class, the tricks here seem to be (1) to understand how people are using the image to sift through information, (2) to eliminate the parts that aren't useful in the given context, and (3) to even rely on the user's imagination/perception to fill in information gaps. To me, perception continues to be one of the most abstract and facinating parts of creating visualizations.
Yuta Morimoto
An sequence of approach of "How to depict all buildings on street" is very interesting. I know making 3D model on certain building require to model its structure, which is almost always consuming time. Recent extensive 3D modeling software such as Maya ease us to make such street view model. Although such superior graphic application demands everyone to acquire expert knowledge and technique, the approach we saw in this class has possibility to achieve easy way for everyone to make 3D street view.
Dmason - Nov 12, 2008 02:33:47 pm
In the lecture, Maneesh referenced Blanz, V., Tarr, M. J., & Bülthoff, H. H. (1999). What object attributes determine canonical views? Perception, 28, 575-600. (http://titan.cog.brown.edu:8080/TarrLab/papers/journal-papers/object-recognition/bltabu99.pdf/view)
I was particularly interested in this paper because I wanted to answer the question if there was a left/right assymmetry, and here is the answer
"Responses are approximately symmetrical. For symmetrical objects, views from the left were selected about as often as views from the right. Individual participants did not consistently select the same side for all of the objects and our data do not indicate an interaction with left- or right-handedness. Still, participants did not simply select sides at random. In debriefing they stated that for individual items they clearly preferred one side to the other"
Also the paper provides an interesting comment regarding the accidental view of the clock which was preferred. As it turns out, while it is easier to IDENTIFY an object with an oblique angle, we tend to IMAGINE them in accidental views, a phenomenon that increases with the visual UTILITY of the object.
Jeff Bowman - Nov 12, 2008 12:25:48 pm
Having read the papers and having seen the lectures, I wonder whether the spread of computer-generated visualization will affect the way we perceive the data. I suppose that looking all the way back to da Vinci's diagrams, structural visualization even predates Playfair's abstract data visualization. However, computer graphics are advancing so fast (Crysis anyone?) that in the matter of decades people could become accustomed enough to certain modes of data visualization.
Does anyone else think that people could get very rapidly accustomed to discerning structure through computer visualization, rather that more typical representations of reality?
Calvin Ardi - Nov 12, 2008 02:51:15 pm
The ghosting issue is one that I had never really thought was that big of a problem; in the given examples on the slides, I didn't question whether the steering column or the battery would exist on the outside of the object. Discussing it in class did bring out some interesting points, such as the ambiguity of where an object is located. It seems like using that technique would require prior familiarity with the object itself. Perhaps an approach could be to ghost the entire object itself, placing emphasis on the inner structure/components.
The example on viewpoint transformations (the street level view was a recent version Grand Theft Auto and the overhead view was taken from a later version Sim City) was also implemented in another game, Streets of Sim City (mentioned briefly in class). Sim City 2000 allowed users to build and design cities to their own liking (overhead view) and other Sim City counterparts allowed users to drive, walk, or fly around (in a helicopter) in the same cities they built (allowing for street level and even overhead views).
HeatherDolan - Nov 13, 2008 011:05:24 pm
I found the Viola and Gröller "Smart Visibility in Visualization" paper a useful history of illustration and explanation of different expressive visualization techniques. Many of the techniques are visually very compelling and exciting to me, but I don't need to extract critical information from them. Many of the examples provided in the paper relate to surgical procedures. Biomedical illustrators, while concerned with aesthetics, also go through extensive training programs that include anatomy and biology. I believe some programs even share small parts of their curriculum with medical schools. While these computer generated approaches to biomedical illustrations leverage existing manual illustration techniques, I wonder if there are efforts to somehow integrate the domain specific knowledge that biomedical illustrators possess into these computer generated illustration techniques. Or has it been found that this isn't necessary? Or is are the techniques/tools intended to be used by specialized illustrators? The paper calls out that computer generated visualizations do not yet rival manually created illustrations, but I felt that they were referring more to expression and aesthetics.
Razvan Carbunescu - Nov 17, 2008 01:56:01 pm
Catching up again with online comments but from this lecture the part that I really liked was the algorithm to create a single long image from a series of different pictures and the way the algorithm automatically chooses the cuts. I've always liked the large panoramic type of pictures and seeing a technique to produce something of the same type 'distortion free' (at least for the viewing plane) was very interesting.
Simon Tan - Nov 18, 2008 12:09:02 am
Following from Jeff's comment, I also felt awe at the fact that Da Vinci was doing exploded views centuries ago. Clearly, we're only scratching the surface of what computer visualization can do. Even if people become 'accustomed' to certain types of visualization (I would argue that they already have; who hasn't seen an exploded view before?), the state of the art could conceivably advance further and deliver new ways of perceiving structure that we haven't thought of yet.
I haven't actually seen too many instances of 'ghosting' in publications or textbook diagrams. Now that I see the apparent problem of discerning relative position, I'm glad it isn't as widely in use as cutaways. The jagged edges that are the hallmark of cutaways are a concrete sign of what is in the front of the image, and I can't imagine ghosting achieving that level of clarity.
