|
| Trip Report |
 |
The Viz 2001 ConferenceSan Diego, California, USAJoanna LengThe Viz 2001 series of conferences is sponsored by the IEEE Computer Society Technical Committee on Visualization and Graphics (http://www.cc.gatech.edu/gvu/tccg/) the third biggest IEEE technical committee. It has run for approximately 10 years and is the biggest Visualization conference of the year. It started as a conference aimed at HPC service providers with many hands on tutorials but has progressively become more oriented towards academic research. It is now a 6 day conference. The first 3 days are for tutorials of which the last 2 days also have symposia running in parallel. There is a well established 2 day symposium on Information Visualization (http://www.infovis.org/infovis2001/) and this year the other symposium was on Parallel and Large Data Visualization (http://waggle.gg.caltech.edu/PVG01/). In fact this year the conference was held in San Diego and one of the co-chairs, Mike Bailey, works at the near by San Diego Supercompter Center (http://www.sdsc.edu/). The final 3 days of the conference was a 3 track event on Scientific Visualization, with a track each for techniques, algorithms and applications. There were also case studies, panels and an exhibition. This conference was too large for one person to attend all the tracks and
sessions. I present here an overview which tries to identify important trends
and techniques. I have also given URL's which may provide information to the
specialist reader.
Parallel TechniquesGraphics hardware is currently the fastest developing hardware compared to processors, hard disc or memory. Until recently the bottleneck in the visualization pipeline has been in the rendering but the improvement in the graphics hardware has pushed the bottleneck higher up the visualization pipeline, for example into the isosurfacer. Many visualization systems have recently come up with strategies for parallelizing particular visualization modules or for streaming the data through the pipeline so that the machines cache is never completely used up. The one day tutorial "Large Scale Data Visualization and Rendering" focused on this topic, there were a strong contingent of VTK users presenting the tutorial and the bias was toward their solution but OpenDX and Chromium were also discussed. There were also papers on this topic included in the "Symposium on Parallel and Large-Data Visualization and Graphics" and in the technical conference. There can never be a completely generic parallel solution for any visualization system. The methodology will favor parallel versions of certain visualization modules. The main concern presented was to produce a visualization that is interactive to a human user. There was an expectation that parallel machines would be used solely for visualization and that CPU cycles would be wasted.
http://www.nas.nasa.gov/Groups/VisTech/
Surface Rendering and Multi-resolution TechniquesAlthough graphics hardware will improve, the size of the data sets will also increase so that at any time it is likely that there will be data sets that are too large for current hardware. There are several types of techniques that aim to improve the speed of rendering. A big topic this year was point rendering which is where a surface is represented as points instead of a triangular mesh commonly used in visualization. The logic being that a point is relatively "light weight". A point can be defined by its 3D coordinates and its colour. A triangular mesh has each apex defined in 3D space with an associated colour but the mesh has extra information, the connectivity between each apex and when the mesh is rendered the colour and light must be interpolated across the surface of each triangle. If there are enough points, then when the camera is far away there will be at least one point for each pixel of the display and the resolution will be high enough for the user to perceive it as a surface. However if there are too few points or if the camera is close gaps will be seen between the points. Some algorithms have been developed to fill in gaps between points for close positioning of the camera. In the tutorial "Multiresolution Techniques for Surfaces and Volumes" the speaker thought the technique would be good for computer games and may soon be incorporated in graphics cards. If objects are defined as meshes in a visualization scene then the level of detail of the mesh needs to be different depending how far the image is from the camera. If the mesh that represents a bunny is far from the camera it may be possible to represent it with 100 triangles but if it is near the camera it may need many more triangles. There are many techniques and method for reducing the size of a mesh while keeping the overall topology of the object. The problem with these techniques is that they are often very application/data dependant and they can not be implemented in hardware. Mesh decimation algorithms tend to be compute intensive and so have been implemented in parallel to make the application more "interactive". They are best not used on the fly but used instead to reduce the complexity of a data set before it is read into a visualization system.
http://www.merl.com/people/pfister/
Flow VisualizationFlow visualization has always been one of the traditional uses of HPC and visualization. There are always a fair number of fluid flow papers and this year was no exception. There was a session on this topic in the Symposium on Parallel and Large-Data Visualization and Graphics. Two of the papers presented useful techniques, "Case Study: Visualizing Ocean Currents With Color and Dithering" and "Real-Time Out-of-Core Visualization of Particle Traces". There were two sessions P2 and P5 on this topic but unfortunately the maths was too complex for me to be able to fully evaluate the methods. However the paper "Quantitative Evaluation of 2D Vector Field Visualization Methods" might be useful for someone wishing to select a suitable method for their 2D data.
Error VisualizationThis topic was not strongly featured in the conference but is of personal interest so I have included a brief description of the topic. Images aid quick analysis and understanding but they are often seen as being absolute or indisputable. How can you be sure that what you see is not an artefact of the visualization technique, features incorrectly classified or the result of "bad" science in the form of in precision, poor analysis or misuse or understanding of parameter space. The importance of this subject was discussed to varying extents in the "Capstone Panel: Battlespace Visualization: A Grand Challenge" from the Symposium of Information Visualization, in the "Panel: Realism, Expressionism, and Abstraction: Applying Art Techniques to Visualization" in the tutorial "Multiresolution Techniques for Surfaces and Volumes" and in one paper "Visualizing 2D Probability Distribution from EOS Satellite Image-Derived Data Sets: A Case Study". Error visualization is very important to the armed forces because important decisions need to be made fast, under stressful conditions, by non-technical personnel and the decisions need to be justifiable/reliable. Unfortunately there is no good visualization solution for error visualization in the armed forces. The members of the panel, who were interested in the use of art in designing visualization, were interested in the use of non-realistic rendering. This is because use of photorealistic rendering makes the user think the visualization is exact, based on fact. Another point made by this panel was that the visualizations that show levels of uncertainty tend to draw the eye to the "uncertain" features that the observer should really try to avoid making decisions from. The tutorial showed a way of displaying two glyphs for maximum and minimum field flow. This looks reasonable but I wonder how easy it is to understand what the maximum and minimum are and how they fit into the rest of parameter space. The paper presented an interesting way of analyzing the data. A set of images were produced that covered all parameter space. Then for each pixel histograms and statistics were performed to show what an average pixel value would be. From analysis of the whole set of statistics it could be seen that one result almost fitted the average value. This was the image they chose to use for further analysis.
Art and VisualizationThere was a panel entitled "Realism, Expressionism, and Abstraction: Applying Art Techniques to Visualization" chaired by Theresa-Marie Rhyne and was the only panel I managed to attend. The panelists showed some interesting work and explained how they had been inspired from art. Each panelist has worked producing visualization for a number of years and has an interest in human perception and how that can be applied to visualization for example to reduce the complexity of images or make them more "intuitive". Theresa-Marie Rhyne http://lts.ncsu.edu/staging/Theresa/tmr.html
The panel ended with a rather heated discussion. I think some of the audience mistakenly thought that the panel were advocating an "artistic" attitude toward visualization where feelings were being expressed not scientific data. The panel saw art as a very structured discipline and thought concepts in composition and image details could be used to produce more "understandable" visualization. In any case it is true that evaluation of a visualization and the metaphors used within is not scientific and is done by user evaluation a method comparable to reviews/critics in art.
Virtual RealityThis visualization community tend to think of VR/VE as large display systems with stereo imaging. I would define VR as systems that try to be immersive and so has many different interfaces which interact with the user by feedback, interfaces could be visual, haptic, sound or any other that has appropriate hardware. However VR can not be reduced to the idea of adding haptic feedback to a system because the visual channel is overloaded. This year there were 4 case study papers in only one VR session and although all four were strong papers only two of these would fit my definition of VR: "Virtual Temporal Bone Dissection: A Case Study" and "A Virtual Environment for Simulated Rat Dissection: A Case Study of Visualization for Astronaut Training". There are other conferences specially for VR but I think in future the overlap of the two fields will increase and the contents of both conferences will change. It may also be that adding computational steering to a simulation may inadvertently produce VR instead of visualization.
Information VisualizationGenerally the scientific visualization community is separate to that of the information visualization community. However there is some overlap. Information visualization tends to work with data bases and produce images related to graphs (often trees or circular displays) although it has been used to produce systems to organize email and debug software. Often techniques used in information visualization are concerned about the way the images are perceived and will enlist psychologists to evaluate techniques. Topics of particular interest in are:
Related Conferences and JournalsThe most important conferences and symposia on scientific visualization are organized by IEEE or Eurographics. As well as the conference series there are also one off symposia which may be of interest these are often advertised through web sites of mailing lists. Other organization may organize useful events such as the EPSRC funded VVEC (Visualization and Virtual Reality Community Club). Many of the large computer societies have useful resources for example ACM SIGGRAPH and SPIE--The International Society for Optical Engineering. Sponsored by The IEEE Technical Committee On Visualization and Graphics - http://www.cc.gatech.edu/gvu/tccg/
Sponsored by Eurographics - http://www.eguk.org.uk/
Sponsored by ACM SIGGRAPH - http://www.siggraph.org/
|
|
|