Conveying Shape:Lines

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Nov 28, 2007



[edit] Readings

  • Automatic illustration of 3D geometric models: Lines. Dooley and Cohen. (acm)
  • Line Direction Matters. Girshick et al. (pdf)
  • Suggestive contours. DeCarlo et al. (html)

Optional Readings

  • Illustrating smooth surfaces. Hertzmann and Zorin. (html)
  • Automatic illustration of 3D geometric models: Surfaces. Dooley and Cohen. (pdf)
  • Speed of Perception as a Function of Mode of Representation. Ryan and Schwartz. (jstor)
  • Assessing the Effect of Non-Photerealistic Rendered images in CAD. Schumann et al. (html)

[edit] Mark Howison - Nov 28, 2007 12:30:11 pm

Here is an article claiming that Cassidy Curtis's drawing is "the best homework ever." Is it also the best tatoo ever?

I wonder how hard it would be to extend Hertzmann and Zorin's algorithm to produce drawings of topological surfaces with "windows" like the ones Curtis uses to show occluded intersection points.

In Maneesh's lecture today, he distinguished photographs from line drawings as being more accurate representations of what we see. I think a more precise statement is that photographs more accurately represent visual stimuli. The problem is that visual perception can be socially constructed, especially among trained professionals (e.g., Goodwin's notion of "professional vision"). For instance, in the experiment with the surgical line drawings, its possible that the semi-schematic drawings more closely represent the highlighted features (tools, incision sites, anatomical structures, etc.) that a surgeon actually "sees" than a photograph would. That is, when the surgeons view the photograph, they actually form a percept more similar (with respect to highlighted features) to the semi-schematic drawing. Meanwhile, a non-surgeon looking at the photograph might not latch onto the same features and may have a very different percept that does not share highlighted features with the semi-schematic drawing.

Another problem with photography is limited exposure latitude, resulting in lower dynamic ranges as compared to what the human eye is capable of. Especially in poor lighting, photographs often don't seem to capture what you "saw" when you took the picture. One way around this is to use high dynamic range imaging, which takes multiple bracketed exposures and tone-maps them using computer software into a single image. Depending on the tone-mapping, this leads to photographs that better capture the highlights and shadows as our eye would see them, or to photographs that look strange and unnatural.

[edit] James Andrews - Dec 02, 2007 09:30:35 pm

The article Mark linked notes: "Visualization techniques have always been important in mathematics and its applications, and they are especially so nowadays as sophisticated computer graphics enhance our ability to interpret phenomena we could not imagine a generation ago. But you can only really appreciate what the computer is showing you if you've tried to render the curves and surfaces freehand."

I wonder what parts of the process of drawing freehand add so much to the understanding of a computer-generated visualization, and if there's any way to animate or allow interactivity to add some of that understanding back to the visualization itself? For example, would it help to give an 'over the shoulder' view of the drawing process, animating the steps used by a human to create the drawing?

[edit] Mcd - Dec 04, 2007 08:27:26 pm

The iNaturalist team was talking before this lecture about the inadequacies of Google Maps for showing features of interest to naturalists (note: they've since added a terrain layer to the API), and I think the work on lines in the direction of principal curvature could be useful in this regard--perhaps even more useful for Wes's project. I wonder if it could be more informative to display topographical information using this approach. It seems that standard topo lines follow one of the principal directions of curvature, but would adding the second aid in the perception of a terrain's shape. The results we saw in yesterday's lecture (the benefits of a late wiki-post) seem to indicate that it would.

[edit] Robin Held - Dec 07, 2007 11:12:24 am

I found the suggestive contours discussion very interesting. I especially like how geometry and a set of mathematical operations can produce something so aesthetically pleasing. It reminded me of cel shading, which has been used in several video games to impart a cartoon style to the graphics:

Note that in the case of cel shading, suggestive contours aren't necessary. This is because flat shading is employed, which provides the shape cues that would otherwise be provided by suggestive contours.

[edit] Amanda Alvarez - Dec 08, 2007 06:29:16 pm

It makes sense that surfaces are perceived better when rendered with lines showing principal directions of curvature. After all, natural surfaces are generally not isotropically textured, and so texturing them anisotropically (not uniformly in all directions) makes sense and is more informative. Also, the first principal direction vectors on a surface have a 'common fate' and thus better convey the impression of a closed continuous surface than a field of random vectors.

[edit] N8agrin - Dec 15, 2007 01:43:00 am

I echo Andrew's sentiments. It seems like the concept of principle line direction as notating shape has long been employed by cartographers. Even simple topographic maps which inherently employee basic principle direction lines seem to provide a sense of shape. I have never come across a map whose topology was defined using principle direction lines explicitly, however it would be interesting to see how this technique could provide dimensionality, particularly in interactive maps.

[edit] Ken-ichi - Dec 15, 2007 05:40:26 pm

In response to the discussion on using the other principle line of curvature for topo maps, I searched around and I think this is similar to cartographic the idea of a hachure, lines that run up and down slopes, sometimes varying in weight to indicate the degree of incline. Occasionally you will see marks like this on maps to indicate low hills, perhaps ones of some significance that don't exceed the contour interval.

I'm not convinced this is a great idea, though. What would these lines tell you about a relatively flat landscape? When things are flat, contour lines disappear, leaving graphical space for other elements. If you were to draw the other line of curvature, would they just be straight lines, or would they end at some contour, like hachures seem to? Also, how would these lines compare to shading techniques? Shading is more popular by far, but is it better?

[add comment]
Personal tools