Trip Report

The Viz 2002 Conference Boston, Massachusetts, USA Joanna Leng The review of last years conference gives more details about this conference and other similar conferences. The conference started in its normal way with 3 days of tutorials/workshops, the last 2 days of which were accompanied by the Symposium on Information Visualization and the Symposium on Volume Visualization and Graphics. The Symposium on Information Visualization is a substantial event in its own right but at its current size and quality it can in no way be considered subordinate to the main conference. The Symposium on Volume Visualization and Graphics runs on alternate years between the Symposium on Parallel and Large Data Visualization. The final 3 days was the main conference which consisted of 3 tracks on Scientific Visualization; a track for techniques, algorithms and applications. There were also case studies, panels, an exhibition and this year for the first time posters. The topics discussed here represent only part of the conference and aim to give an overview of currently important themes. The organisation of the conference makes it appear that information visualization and scientific visualization are totally different disciplines but there is not completely true. A simple explanation of the difference is that scientific visualization tends to use data that lies on a 2D or 3D geometry for example a medical scan or a computational simulation of air flow over an aeroplane. While information visualization tends to use data that is quite abstract, may be dependant on many variables and is not associated to a geometry or location for example the contents of a file system or economic/stock market data. However the overlap consists of examples like understanding telephone call patterns across a country or understanding the interaction of different parameters used in a simulation. There was talk at the conference of splitting the Symposium on Information Visualization from the main conference but to me that would be a mistake. There are many core technologies and approaches appropriate to both disciplines and currently researchers interested in both disciplines can attend just one event (with one set of travel costs) while researchers who have a strong bias may attend just one half of the event. A keenly debated topic was how to improve the take up of visualization in academia, medicine and industry. As this is an important question I have written about it here extensively. Visualization is an applied science like statistics, computation or graphic design. It can not sensibly exist as a subject on its own, it needs data and people who are interested in that data. The discussion centred around what roles we should expect of different experts. Should a computer scientists just design a system and then move onto another project? How generic can a design be especially in an evolving domain area? Domain experts must participate in the design process but at what point can/should they take over? Two panel sessions "Volume Rendering in Medical Applications: We've got pretty images, what's left to do?" and the "Visualization in Bioinformatics and Cheminformatics" centred on these issues. Bioinformatics is new to visualization and there are many visualization tools and techniques that need to be developed. Nevertheless there are some well known tools like spotfire (http://www.spotfire.com/) which use visualization to support decision making particularly in bioinformatics. Anecdotal evidence indicated that even when visualization tools were available many bioinformatics experts preferred to continue using excel. Some thought the users were keen on new technology and just needed/wanted training while others thought they were too overstretched to find the time/capacity to learn new technology. By contrast techniques for 3D medical visualization are mature but the medical profession have not taken to them. The medical profession is generally an overworked and stressed group. They tend to take up new technology only when the benefit is obvious: 1, it speeds up diagnosis 2, it allows the diagnosis/treatment of a condition that could not previously be diagnosed/treated. It is interesting to note that in the USA it is private medical scanning services that are making use of 3D techniques because the patients like it. In both of these cases it appears that the visualization community has a body of knowledge that does not seem to be moving into the general user community. Knowledge flow is slow. The younger Nintendo generation are happier with computers/visualization than their predecessors but how can we speed up the process? Should introductory visualization become a subject taught to all science students like statistics is taught now? Should the visualization expert work more closely with the domain expert? Should computer scientists be concerned whether their system is eventually used or is it better to flood the field with solutions until a preferred techniques are selected? Is there reward for collaboration in academia? Large Data, Parallel Methods and out of Core Methods The subject of dealing with large data was by far the biggest topic of the whole conference. This theme was divided into several distinct areas including out of core algorithms, compression of meshes, simplification of meshes and efficient rendering. Although graphics hardware is better than ever before the size of the data has increased. Techniques dealing with large data evolve to deal with new hardware and types of data. The half day tutorial called "Out of Core Algorithms for Scientific Vis and Computer Graphics" gave a thorough overview of algorithms for large data. Out of core algorithms are designed to handle datasets larger than main memory. There are a class of algorithm that presort the data using a tree like structure that speeds up the display of data. These trees can be used to structure the data so that for example relative few data points need to be searched to find an isosurface or so that it is easy to select points on a surface that face the camera for view dependant algorithms. When the tree becomes too big to fit in memory then the out of core or cache friendly algorithms become important because they use disc based data structures that allow data to be paged to memory efficiently. "Batched" or "streamed" computations are a class of out of core algorithm where the data is divided into small portions that fit in main memory, are passed through the visualization pipeline and a final image is stitched together on the screen or in an image file. The details of the papers on the general theme of dealing with large data are not suitable for this report. Understanding these algorithms requires too much specialist knowledge of visualization algorithms and particular parallel programming techniques. This report is aimed at giving an overview of currently important areas in visualization. VolVis Important Papers Volume visualization consists of techniques used to display and classify the important features in a 3D block of data. A technique is a method that can be applied to data from one domain area, an application is a technique which can be used on data from a number of domain areas while a toolkit or case study consists of a number techniques used for analysis of a certain type of data. Many volume visualization techniques presented at the conference were for large data few new techniques were presented for small (data of any size), the most interesting were in the sessions "Level Sets and Isovalues". Two of the three papers from this session presented techniques that aimed to find "meaningful" isosurfaces from the a 3D block of data. "Exploring Scalar Fields Using Critical Isovalues" presented a technique for finding a series of isovalues that produce isosurfaces with different topologies. For example this would be compute the change in an isosurface from being a continuous surface to one with a hole. While "Efficient Computation of the Topology of Level Sets" used a Contour Tree to determine the number of isosurface components there are for any one isovalue. The tree allows a user knows to the number of individual surfaces there are and if some are not visible the user knows to explore the data because some surfaces are hidden within others. A Contour Tree is a structure commonly used to preprocessing of data so isosurfaces can by more quickly extracted. Assessing a Visualization The ability to assess a visualization and determine how good it is a difficult problem and one which reflects the complexity and multi disciplinary nature of visualization. Should a visualization be assessed by: how well it compares to reality how easy it is to understand how well it adheres to simple rules of perception and design if it shows something previously not seen This year it was the subject of many talks and tutorials. The tutorial "Information Visualization, Visual Data Mining, and Its Application to Drug Design" by Georges Grinstein of the University of Massachusetts at Lowell and Daniel Kein of University Constance gave a comprehensive talk on techniques in information visualization. The talk covered many techniques and paid particular attention to perception. Their primary concern was to increase the speed that a person understood a visualization or parts of it. With information visualization one of the aims is to cross correlate as many parameters as possible. This means as little screen space as possible must be given to each parameter and the quicker the image is understood the more information can be compared. The Capstone talk for the Symposium on Information Visualization on "Display Design for the Eye and Mind" by Stephen Kosslyn of Harvard University. Stephen Kosslyn is a psychologist working in information visualization. He has listed the key perceptual features that are important to know so that a designer can make important information stand out while unimportant information does not. Generally speaking important features of a visualization should be quickly identified. Fast perceptual understanding is "hardwired" into the brain in what are known as preattentive channels. These channels are said to aid survival by either recognising danger and/or the lack of it, for example spotting a predator. Confusing a preattentive channel will distract the viewer and should be avoided for example an object illuminated from underneath will confuse. The tutorial "Pschychometrics 101: How to Design, Conduct, and Analyze Perceptual Studies of Computer Graphics Visualization Techniques" was given by James Ferwerda from Cornell University, Holly Rushmeier from IBM and Benjamin Watson from Northwestern University. Pschychometrics use psychological experiments to prove how "real" a computer graphics scene is or hoe quickly/accurately information is interpreted from a image. None of these talks address the problem of how to encapsulate error in a visualization but to me it is key to being able to assess a visualization. Error estimation remains an important but relatively unexplored area of visualization. Display Technologies Display technologies are important to all forms of visualization, VR or any application that is visually displayed. The panel "Combining Sensory Information to Improve Visualization" covered many areas of display technology. Martin Banks from the Visual Space Perception Laboratory at UC Berkeley talked about how flat screens hinder the understanding of depth, 3D form and orientation of an object. He presented an intersting display devise that used 3 partially-silvered mirrors to present the eye with a multi-focal display. Tests showed this gave better depth perception than a flat screen. See http://john.berkeley.edu/. Felix Wichmann from the Max-Planck-Institute for Biological Cybernetics discussed the limitations of the CRT displays. Many visualizations use luminance to indicate the strength of a parameter but with a CRT the luminance is dependant on the voltage not the image. Medical images are in gray scale and the effects of luminance mean more detail is lost than in a colour image. Images are distorted differently in horizontal to the vertical, a deliberate artifact used to make text more crisp. The presenter described a simple experiment to allow you to try and understand the distortion. Take an image on paper and view it. At the same time take a video of the image and distort it until the 2 look the same. Laurence Maloney from New York University talked about the lighting model and how it should be adapted to give spectrally-accurate rendering. Improving the lighting model is still one of the big challenges for computer graphics. Finally Heinrich Bulthoff the director of the department of cognitive and computational psychophysics at the Max-Planck-Institute for Biological Cybernetics gave an interesting overview of pschycophysics. His department has many types of virtual environment that they use study questions like which has the biggest impact on perception stereo, motion or colour; how good are virtual environments for training; how important are local and global cues and how robust is spatial knowledge. There were also two papers that explored the use of display technology. The paper "'Case Study: The office of Real Soon Now' for Visualization" was presented by Samuel Uselton from the Lawrence Livermore National Laboratory. This case study showed how they have fitted two offices with two projectors so one wall was used for display rather than a screen. It has always been difficult to get people to use large scale display environments although they are useful especially for collaborative work. The two people who used the system like having a large display area always there and found discussions with colleagues became more productive. As projectors become less expensive it is likely that this may well become more common. A Case Study on the "Applications of a Generic Library for Low-Cost Polychromatic Passive Stereo" presented a preload library which enables most OpenGL based applications to display high-quality polychromatic passive stereo. This solution was designed to be portable in terms of hardware, software (the preload libraries allows this to work with software for which no source code is provided) and display devise. Although this application requires users to ware glasses, the ability to see in stereo when desired would be incredibly useful. Haptics The difference between visualization and VR is that visualization allows users to explore digital data while VR immerses a person in an environment that often mimics reality. This tends to mean visualization is limited to visual display with occasional use of sound or haptics when the visual channel is overloaded while VR uses display and feedback devises to make the user feel they are "inside" the data. This distinction is normally reflected in the conference by the number and type of papers there are on feedback devises. This year there were two papers on haptics as part of the symposium on volume visualization. Both were based on work done on a phantom. A motorised devise that consists of a pen like structure that moves with 3 degrees of freedom. The paper "Haptics-based Volumetric Modelling Using Dynamic Spline-based Implicit Functions" looked at modelling geometric objects within a computer graphics system. Producing a mesh appropriate for simulation or as even as part of a virtual reality or visualization has always been problematic, it is one of the big problems for finite element analysis and for medical training environments. While the paper "Real-time Volume Haptic Rendering with Shape-Retaining Chain Mail Model" displayed work done on speeding up the chain mail model, a model originally designed to have similar properties to human soft tissues. This work would be useful for medical simulations or training tools.