From CS294-10 Visualization Sp11
Visualization of Digital Photograph EXIF Data
I want to create visualizations exploring the wealth of information contained in the EXIF data of digital photographs. The prevalence and sophistication of digital image metadata has increased substantially since it was introduced 10 years ago. Camera exposure settings, equipment data, and even geotagging information are now commonly attached to photos shared on the internet. This wealth of metadata is largely untapped and can be incredibly instructive for individual photographers curious to learn about the shooting techniques of other photographers. Visualizing EXIF data can also help photographers better understand their own shooting behavior. Currently, it is common practice to study the EXIF info for individual photos that are interesting, but the practice is slow and laborious. Aggregate analysis can rapidly reveal shooting trends that can otherwise be hard to identify.
Visualization of image metadata facilitates comparison tasks, for example, between similar images taken using different settings. These comparisons reveal the consequences of adjusting various exposure levers on the resulting photograph. This analysis is not limited to factors concerning exposure, but includes variables recorded in the EXIF info, including metering modes, flash on/off, derived measures of depth of field, and so on. The user might input a collection of photographs deemed high quality of a specific subject matter, such as landscapes, and see how the photos are distributed on a plot in the space of aperture vs. focal length. The user might gather best practices and a better understanding of how these settings affect landscape photographs. Reference images are abundant, with sites like Flickr offering easy access to huge repositories of user-tagged/rated content.
Geotagging has recently become popular with photographers, made possible by GPS enabled cameras. Systems such as Google Earth have taken advantage of the explosion of geotagged images, attaching photos to interactive maps. Crunching geotag data can produce statistics such as the geographic distribution of owners of different camera makes. But truly interesting visualizations are possible by combining geographic information with time and date. For example, the additional dimensions of season or year can be factored into a visualization of geographically arranged photographs. A browsing user might be able to explore how the Pacific sunrise, or NY's Central Park changes month by month, season by season via the photographs of others.
Link to slides: File:Photovis.pdf
Jvoytek - Apr 11, 2011 03:25:11 pm
This sounds like a really nice tool for professional or "pro-sumer" photographers. Seems like there is a lot of complexity to work out, good luck!
Siamak Faridani - Apr 11, 2011 03:33:49 pm
I think it is an interesting idea. It seems that you are well aware of what is available but I think it might be beneficial to look a little more into hobby projects. There are a lot of projects just around flickr so it won't hurt to look into what is missing and try to come up with features that bridge these gaps.
I am also interested in knowing if you have thought about visualization as an educational tool. For example I can look into the meta data for a photo, learn it and perhaps load it into my camera.
Dan - Apr 11, 2011 03:20:44 pm
This idea seems very practical, EXIF data is used by everyone who uses a camera. You brought up some interesting use cases, such as utilization of equipment use and lens, I think that is important to tell a story so people understand how to use the visualization. The interface also looks like you put a good amount of thought into it.
David Wong - Apr 11, 2011 03:37:13 pm
Great idea! It'd be good to identify whether you're catering this to experienced or inexperienced photographs. As per Maneesh's comment, brushing and linking specifically in the context of photographs could be great to illustrate how certain settings change the overall photograph and be very helpful to a novice photographer trying to learn how to tune their camera. In the same fashion, a specific set of queries could help experienced photographers discover new radical ways of taking photos that they haven't yet tried.
Matthew Can - Apr 11, 2011 06:17:36 pm
I think you did a nice job communicating the kind of tool you plan to build. The storyboard was well done. You would probably get the best feedback for this project from the people that are most likely to use it. I would iterate by doing a pilot study with professional photographers.
Brandon Liu - Apr 11, 2011 08:05:09 pm
I stumbled across a related paper that uses geolocated picture data to create some interesting visualizations: http://www.cs.cornell.edu/home/kleinber/www09-photos.pdf
Sally Ahn - Apr 11, 2011 11:28:00 pm
I think the disconnect you showed between the graphs for this data and the photographs themselves motivates the problem well. People of all skill levels can distinguish "good" photos from the "bad," but it takes skill and experience to interpret EXIF data. For example, I had no knowledge of EXIF data, so the graphs meant nothing to me, but if there had been clear mappings between the graphs and visible photographs, I may have grasped a better understanding of the data. As suggested in class, user testing would be a great way to pinpoint on the particular types of stories you want your visualization to tell about this data. However, I don't think you need to limit your users to professional photographers, since this tool sounds like it could be very useful for novice photographers who want to learn more.
Michael Cohen - Apr 12, 2011 01:28:41 am
This is probably out of scope for your project, but it might be interesting to work in some of the things that David is looking at. For instance, the ability to collapse all the EXIF data (or some parts of it) down to a couple of dimensions and plot them in space to get a general sense of what pictures were taken with similar and different techniques. There could then be an option to show pictures that have similar (or very different) settings as compared to an example or set of examples.
Krishna - Apr 12, 2011 12:19:54 pm
Very nice project, in your storyboard I remember seeing barcharts when a subset of images are selected, maybe you may want to include timeseries plots as well so that users can visualize correlations between the numerical fields in EXIF. An interesting extension of your project would be to visualize correlations between words in the Flickr comments and EXIF values ..
Saung Li - Apr 12, 2011 06:49:09 pm
This looks like a great project, and from your storyboards I can see where your going. Adding in techniques such as generalized selection and zooming will be important for dealing with areas of clutter, especially in your photoplot sketch. Have you thought about adding in location as a variable? This may be too much to do for this one project, but it may be an interesting dimension to look at if you'd like to look for certain correlations.
Julian Limon - Apr 12, 2011 10:42:07 pm
This looks like a very interesting problem. I agree with Michael Cohen in the sense that some dimensionality reduction techniques could be used to organize collections more effectively, since a large number of photos may have been taken with similar exposure. This may not be as useful for expert photographers but could be interesting for novice photographers who have downloaded a set of pictures that they like and want to learn more about.
One other possible idea would be to allow multiple people to collaborate on a set of pictures. Groups of experts may derive some patterns from the data that may be useful for others who are also interested in these photographs.