Conveying Structure

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Nov 14, 2007

Slides

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)

Contents

[edit] Ken-ichi - Nov 12, 2007 03:46:42 pm

On McGuffin et al.

First of all, wow, making tools called "The Peeler" and "The Hinge Spreader" for cutting apart a human head, even a digital model of a head, is kind of freaky. That said, it is also kind of awesome. I really like the cutting hinge, because the hinge/book metaphor seems so natural. You could imagine each layer exposing a seam that you could crack open like a book. I think the distortion of the opened layer far from the seam could be problematic, but I'm guessing that the region of focus will be the area around the seam, so maybe not. My favorite is probably the leafer tool, because I feel the opening box metaphor is still pretty natural, but original shapes are preserved.

[edit] Willettw - Nov 13, 2007 12:51:29 pm

Reading through these, I remembered another interesting paper from this year's Vis conference that might be of interest.

Illustrative Deformation for Data Exploration (Correa, Silver, Chen)

Here, the authors apply a number of deformation techniques similar to those in the McGuffin paper to a more general set of 3D data visualizations under the auspices of data exploration. For example, in one case they use deformation to split and refine a selection in a 3D scatter plot and in another they distort and slice video volumes over time for analysis.

The idea of using these sorts of distortions on visualizations that aren't actually representations of 3D objects (unlike most of the papers above) is interesting, but also seems potentially problematic to me in a lot of cases since doing so may misrepresent the data. Still, as tools for interactive analysis and exploration of 3D visualizations some of these techniques do seem promising.

[edit] David Jacobs - Nov 14, 2007 01:23:56 pm

On the subject of cutaway models, there's a pretty cool one of an electric motor on the 1st floor of Cory Hall. It's in the north-west corner of the building right outside the CS 150 Lab (125 Cory for all you grad students, which is pretty much everyone).

[edit] Hazel Onsrud - Nov 14, 2007 07:33:10 pm

The study by Palmer, Rosch and Chase examined in class, found that their participants tended to prefer a frontal view from above, approx. 3/4 up, however there were some exceptions. Noting some of the exceptions, including the house and (arguably) the horse, I wonder if the subjects tended to pick different views for these objects because our normal perspective of them is different. For example, we don't often look down at either object. I wonder what a similar study might show if the preferences of small children or short adults where compared to tall adults.

[edit] N8agrin - Nov 19, 2007 10:11:41 am

Comparing McGuffin et al.'s approach of volumetric data browsing to Li et al.'s interactive exploded diagrams highlight the differences and difficulties involved in manipulating structural information between 3d and 2.5d information.

In inherently 2d information, determining the correct occlusion for individual pieces seems like one of the more difficult tasks to automate. This task is made simpler by allowing the pieces to themselves be broken down into fragments. On the other hand, using pure 3d data occlusion can be defined dynamically simply through the dimensions of each individual 3-dimensional subcomponent. Still, as McGuffin shows, manipulating the 3d information to show into models requires gracefully deforming the data uniformly.

Having worked in an engineering firm building tangible objects from 2.5d schematics, I appreciate Li et al.'s work and can see its utility. Still, I'm curious as to the frequency, in today's age of pervasive CAD based design, how frequently 3 dimensional models aren't available to create the same sort of 2.5d visualizations. Still, I see how avoiding the manipulation of 3 dimensional data can be important in certain environments, like vector based animations deployed via the internet where minimizing data delivery is paramount.

[edit] Robin Held - Nov 26, 2007 02:52:09 pm

The article by McGuffin et al. touches on the facts that any visualization technique designed for radiologists or other physicians must have a very low learning curve and have very clear benefits over existing solutions. I have interviewed radiologists for previous class work and research projects, and those two items are universal. It boils down to radiologists already having a wealth of tools at their disposal and being resistant to abandon familiar routines, even if they are not as efficient as the latest technology. Therefore, it is crucial for groups like McGuffin et al to perfect the cursors their software offers for navigation. If it's cumbersome or nonintuitive, it will be extremely difficult to convince medical practitioners to give it much thought.

[edit] Amanda Alvarez - Nov 28, 2007 06:07:13 pm

The general theme of the readings and the lecture was extending the viewable space beyond that which is specified by simple light propagation (which leads to opacity and occlusion); this extension is achieved through distortions, cut-aways, transparency, etc., to allow for abstractions of the object's structure to become present and tangible. This makes me wonder whether the user is forced to work harder to understand the counterintuitive display; or does it liberate the viewer from a constrained viewing situation that shows only the outside of an object? The visualization after all is supposed to simplify, uncomplicate, and bring the key features to the fore, whereas the visibility manipulations require detailed animated transitions and cluttering widgets to keep the user on board with all the transformations.

McGuffin et al. allowed users to make physically impossible cuts and deformations in volumetric data. If these manipulations cannot be physically realized, what exactly is it that one is visualizing? A cross between reality and some guesstimation of how the layers behave during deformation? If you can't peel away bone easily in the real world, how is it informative to be able to peel away bone in the visualization? When do the animated transitions become more representative of fantasy than reality? In some respects, that which makes smart visibility 'smart' can also be a weakness; in trying to peel away the outside to get to the hidden inside, we may be making up that inside, or at the very least misrepresenting it (which is the whole point of the process in the first place).

As McGuffin et al. point out, layers may not be the most intuitive way to think of volumetric data. I don't really have a clear understanding of how layers are specified; seems like it has to be done manually for each new dataset, unless there happens to be a very well defined class that always has non-varying layers of interest (ie. not the human body). Once layers are specified, it constrains the possible manipulations of the vis quite a lot. Skin, muscle, and bone may be too coarse of a layering classification for some purposes, which would then require new layer definitions.



[add comment]
Personal tools