Using Space Effectively: 2D II

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Oct 5, 2007



[edit] Readings

  • A general cartographic labeling algorithm, Edmondson, Marks & Shieber (pdf)
  • A survey of automated layout techniques for information presentations, Lok and Feiner (pdf)
  • Dynamic space management for user interfaces, Bell & Feiner (pdf)

Optional Readings

  • Rendering effective routemaps, Agrawala & Stolte (pdf)
  • Artistic resizing, Dragicevic et al. (pdf)
  • Adaptive grid-based document layout, Jacobs et al. (acm)
  • Map labeling bibliography


[edit] N8agrin - Oct 02, 2007 11:15:53 pm

Each reading discussed a particular method for handling the automatic layout of information in an interface or visualization. I was a bit alarmed by the Dynamic Space Management for Interfaces paper, and had a difficult time coming up with any practical uses for this sort of algorithm that wouldn't violate general usability principles. The exception to this concern was the author's comment on the possibility of using the proposed algorithm to handle batch adding of user interface windows. The only other applications I could think of were for certain specialized environments like airplane cockpits or other fast paced, critical decision making environments, where preventing new windows of information from overlapping might be useful.

The discussion of automated layout techniques for information presentations relates somewhat to other research I'm involved in, focused on building abstract definitions of user interfaces, and having to reinforce the need for components of the interface to relate to one another via constraints imposed upon them by an underlying data model.

The reading on the layout of cartographic information seemed the most obviously applicable, especially in today's age of electronic maps, gps and navigation systems. Aside from these applications of the technique, the results were beautiful and visually demonstrated the power of their algorithmic approach. For what it's worth, I also appreciated their humble attitude towards their approach and their insistence that a computer may not always be capable of deciding how to best layout information that, under specific constraints, becomes a purely aesthetic assessment of the overall layout. I thought their approach particularly elegant as well, using annealing theory, something I more closely relate to DNA and RNA behavior than computer science, to accommodate what might otherwise be considered an artistic skill.

[edit] Kenghao Chang - Oct 03, 2007 08:48:51 am

From Marks & Shieber's map labeling paper, we know map labeling is actually a constraint satisfaction problem; a label should be closed to the target and apart from other labels to be legible. In Figure 10 of the paper, it shows a map with tons of labels on it in order to demonstrate the effectiveness of its algorithm. If we look into the rendered map, there is little confusion between any two closed targets and their labels. However, the overall picture is overwhelming: there are too many labels in a map (labels are too dense). This reminds me that many exisiting map software, such as google maps, filter out details in order to make the overall layout clean. As users zoom in, the details will be shown. As a result, I believe the upper limit for density of labels should be considered as another constraint.

[edit] Andrew McDiarmid - Oct 03, 2007 09:08:48 am

Reading these papers on algorithmic layouts reminded me of something I learned as part of my first job at a small art museum, where one of my duties was to assist in editing labels and in-house publications. The editor taught me that always following grammatical and graphical rules is not as important as internal consistency, which in turn is not as valuable as judiciously violating that consistency. In short, publications communicate best when an editor understands the guidelines well enough to know when to break them, and I think that is something easily lost if layout and design are left entirely to algorithms.

The previous comment noted that the result of Marks & Sheiber's map-labeler was overwhelming, and could have benefited from pruning. Rather than adding an additional constraint as Kenghao Chang suggests, I wonder about a visualization system that outputs not a static final image, but perhaps a scalable vector graphics file in which non-data such as labels and graphical layout (to prevent distortion of data, quantitative and scaled images would not be editable except at a very high level) could be hand-edited to adjust what is communicated to a specific task (e.g. publication).

[edit] David Purdy - Oct 03, 2007 11:42:17 am

Several examples of election cartograms are at:

[edit] Ariel Rokem - Oct 03, 2007 02:00:54 pm

The discussion of zooming and distortion in layout reminded me of a talk I heard by Eli Peli [1] a couple of weeks ago. Peli and his colleagues are working on a system to help people with loss of the peripheral visual field. They found that instead of a lens that collapses a lot of the visual field into their limited visual field (essentially zooming out), these people prefer a system that provides them with two images of two different scales simultaneously overlaid on top of each other. Thus, they prefer to retain their limited central visual field, in addition to a wire-frame rendering of the edges existing in a field of view larger than their own visual field. I believe that a similar trick can possibly be used in order to show data at two different scales simultaneously. One scale will be shown in one rendering and the other scale (zoomed out, for example) can then be shown with a different, lighter rendering. The proof, as usual, is in the pudding.

[edit] Jimmy - Oct 04, 2007 09:54:17 am

Kenghao, the google map is also a good application of semantic zooming, which adds more details when approaching a particular area of the graphic. When zooming into a certain area of the map, more detailed street information will show up.

The Marks & Shieber paper points out a problem of zoom and pan. When the user zooms into a certain area and wants to see another area, she needs to first zoom out, pan to another place, then zoom in again. The online map services have managed to deal with this problem. For Google map or Yahoo map, there's a small map beside the main map. The small map is an overview that covers larger area (zoomed out map). So the user can more quickly view different locations using this small map to pan to other places without doing zooming out/in again.

In the last part of the lecture we see the London Subway system map that is the distorted format of the actual map. Such distortion reduces the complexity of the original geographic layout and clearly displays the stations and lines. As we don't care about the actual geographic layout, it's fine for the distortion. However, the BART system map keeps the geographic layout and still doing well to visualize the map:


Probably it's because the BART system is a lot less complicated, with only a few lines in the Bay Area.

[edit] Robin Held - Oct 04, 2007 02:57:32 pm

After reading the paper by Edmondson et al, I have a bit of trouble believing it would be universally desirable to have a COMPLETELY automated layout system for geographic maps, with no room for navigation or adjustment. I realize that automation would be useful for dynamically viewing different data sets using a given layout type, but when we're dealing with a topic that relies on very subjective evaluations, it seems unwise to remove the user from the process. The paper by Edmondson et al certainly presents some impressive maps with efficient labels, but I think an optimal system would include a stage where a user can manually adjust certain labels. The system could then take that input as a constraint, and perhaps readjust the rest of the map in response. The ability of prefuse to use spring-force relationships to set up nodes is a good example of such an implementation. So while I believe that the automation research is interesting and could be very useful, I also believe that we should remember that sometimes a human element must be included to ensure some creative breathing room.

[edit] Wes Willett - Oct 04, 2007 11:31:14 am

While we're talking about label layout, this new paper: Specifying Label Layout Styles by Example (on which Maneesh is actually a co-author) is definitely interesting. Maneesh may mention it in the next class, (or there may be a specific reason for not mentioning it) but it seems to be a fairly nice bridge between two classes of layout techniques - constraints-based layout and learning techniques - presented in the Lok and Feiner survey. The approach taken in the paper allows designers to define layout constraints by example. After a designer lays out labels on an exemplar diagram the system uses machine learning to determine the terms in an energy function that defines how the labels are laid out. This energy function is then used to perform layout on other diagrams, hopefully maintaining the same style. It doesn't work in all cases (when there aren't enough labels in the exemplar, for example), and I wonder a little bit if their set of energy terms encompasses the entire space of layouts a designer might want to produce, but the result is definitely neat.

[edit] David Jacobs - Oct 08, 2007 10:50:52 am

Here's a working link to the "Powers of Ten" video. One of the things I think would be interesting to add to this visualization would be a velocity display. From looking at the video, it appears to be zooming in and out at a constant rate, but it'd have to be accelerating in some interesting way. Especially because we probably spend some time going faster than the speed of light towards the 100 million lightyears range.

[edit] Mark Howison - Oct 08, 2007 02:17:47 pm

David: Actually, the speed is increasingly exponentially, since it is the power of 10 that is increasing in constant time every 10 seconds. You are right that the camera is traveling faster than the speed of light during part of the movie. In the frame encompassing 100 square light years, the x and y dimensions have just increased by 10 light years over 10 seconds, for an average speed of 1 lightyear per second, which is 31536000 times the speed of light (i.e. the number of seconds in a year). I agree that adding a speedometer would aid the interpretation of the visualization.

[edit] Amanda Alvarez - Oct 09, 2007 08:14:11 pm

The algorithm of Edmonson et al. is an elegant way to achieve global robustness of layout; as the name suggests, annealing does not lead to a brittle end result. They also avoid having to explicitly articulate and obtain constraints. The paper mentions that additional expressivity should not be a problem for their approach, but it is debatable whether they would be able to tackle more abstract constraints (in maps of course spatial constrainsts are most important).

I liked the suggestion of Lok and Feiner (at the very end) of incorporating constraints that originate with the user, eg. eye tracking, viewing distance. This could have (has had) tremendous positive implications for layout and usability.

I didn't know that the original idea for Powers of Ten came from this book by Kees Boeke (Cosmic View, the Universe in 40 Jumps):

The Eames clip came a decade later.

[edit] James Andrews - Oct 09, 2007 11:10:51 pm

Jimmy -- for the problem of panning on a zoomed view, I think Takeo Igarashi's "Speed-dependent Automatic Zooming" offers a pretty neat solution. Since you cannot process the fine details when panning, it automatically zooms out an amount proportional to the speed of your pan. There's a simple demo of it ( here: ) so you can get a sense of how well it works.

[edit] Omar - Oct 10, 2007 10:42:23 am

in class maneesh spoke about nomograms. maneesh and others commented that nomograms might give more intuition about the topic of inquiry than direct calculation. this may be true, but then the question is whether this intuition is valuable. here's an example that i think is very interesting:

i was driving in the alameda area with a tomtom (a gps navigation device). it got me from point A to point B, no problem, but on reflection, i have no sense of alameda, the shape or the layout. the display was all focus, very little context. is this a good or bad thing? my friend brought this up, and we discussed that it's likely a fine thing when were traveling, as do we really care about the context of the map -- will we be either without the gps device, or needing an intuition of the space? probably not. but, in terms of alameda, i may go back someitme, likely without the gps device, and i probably want to thus have a sense of its layout.

so its interesting that the map, thought of as a nomogram (i'm doing a path calculation), not only gets us from point A to point B, but also teaches us a lot about the space. focus turn-by-turn maps, like on GPS devices, are excellent at getting me from A to B, but do not help me learn anything.

[edit] Ken-ichi - Oct 18, 2007 09:21:19 am

Omar, is a map a nomogram? I guess it depends on the projection, what kind of spatial relationships it preserves (area vs. distance), and what kind of things you want to calculate. I really like the idea of nomograms, but I'm having trouble thinking of complex calculations that I do on a regular basis that might necessitate them. Maybe calculating tips at restaurants. One interesting example I found fishing around was a nomogram for calculating the Richter magnitude of an earthquake at a particular location given the amplitude of the wave measured at that location and the distance from the epicenter.

[edit] Hazel Onsrud - Nov 05, 2007 05:28:30 pm

I second Kenichi's comment: "I really like the idea of nomograms [being able to spatially view mathematical relationships in this way fantastic for improving the comprehension of spatially-oriented people], but I'm having trouble thinking of complex calculations that I do on a regular basis that might necessitate them."

However, I found at least one example that is still commonly used. A doctor recently showed me a Body Mass Index Nomogram: I thought it nicely illustrated the concept and can see why it is still used.

On the other hand, most of the other nomograms I've come across have not been as useful. For example, while an Acetaminophen Toxicity Nomogram ( displays useful data, I felt it was more likely that this general information would be stored and accessed in a person's brain, and that if specific numbers were needed for more particular calculations, a computer would be used.

[add comment]
Personal tools