Interested in Scientific Visualization? Want to find out more about it as a system developer or a real user with real applications? Then please see the section on forthcoming books.
Norman Wiseman invites comments on his proposal for a Workshop on Graphics Support for the 1990s. This is in response to a perceived need amongst computer service staff that they may need to adapt the way they provide services to the user in the light of the changing nature of computing: distributed, more products, more vertical solutions, changing customer needs.
There are two reports on SIGGRAPH '91. One from a European perspective by Ken Brodlie and the other from a North American perspective by Brian Wyvill. As SIGGRAPH is a major event, readers will probably appreciate this information. We thank both contributors who gave their articles within about three days of the request.
Bob McGonigle continues the section on Graphics around the Country with contributions on the National Microform service at the University of London Computer Centre (ULCC).
The AGOCG Meeting on 9 July 1991 was a full one with over 30 papers to be discussed We were just showing off to Anne Mumford, our Coordinator, that some progress could still be made without her cracking the whip on a daily basis. Anne, who is still convalescing, came to lunch and we hope she will be back to full fitness by now.
On the GKS front, the PC version is nearly ready for Beta testing and we expect a site licence being available from CHEST in the Autumn. The GKS-3D being developed at Manchester is nearly complete and compares quite favourably in terms of functionality and robustness with the GTS-GRAL version. The merge with the validated GKS-2D code is about to start.
The PHIGS benchmark suite is progressing well and the plan is to start the benchmarking of SUN, IBM, DEC, HP, Silicon Graphics and possibly other offerings in September. We shall also test the FIGARO version on a number of platforms. We hope that the reports of this assessment will be available before Christmas. Believing that the use of PHIGS is on the increase, we are establishing an e-mail advisory service so that queries on a number of systems can be answered.
The CGM toolkit Release 1.20 is due next month and we have a set of enhancements planned until next May which will add new functionality, provide support for CALS profiling and provide additional output drivers including a skeleton driver for people to use.
To enhance the colour printer deal, we are going to develop a colour postscript previewer which should help to keep the consumable costs down.
The Visualisation Workshop has produced its two Springer books, an Introductory Guide and the full Workshop report. Both should be on the streets in time for Visualization 91 in October. Much effort was put in by a whole set of people with very worthwhile results.
Two new assessments are just starting, one for high-line rate video projectors and another for colour scanners. It is hoped to complete both before October.
It has been a busy period. Hopefully, we can give most of the work back to Anne to do next quarter!
Edinburgh University Computing Service (EUCS) took a policy decision some three years or so ago to standardise on GKS as the main supported graphics subroutine library for 2-D applications. An earlier survey of microcomputer graphics programming applications in the EUCS user community had revealed a plethora of graphics libraries in use, the range and diversity of which would have been impossible to support in any sensible manner. However, it was clear from the survey that many applications used a particular graphics library, not necessarily from any sort of considered choice, but often as a direct consequence of a lack of choice on the particular computer system being used. A significant number of the applications could have used GKS if a suitable implementation had been available. With increasing standardisation on the use of IBM PCs and compatibles in applications areas where graphics programming was likely to be a requirement, and with an ever-growing number of Unix based workstations, coupled with the University's strategic decision to move towards a distributed computing environment in which Unix would play a significant role, the availability of both an MS-DOS and a Unix based GKS at a minimal level of 2b was essential if EUCS support for graphics programming was to be capable of being distributed across a wide range of systems, from the IBM PC at one end, to Unix mainframe hosts and parallel computing systems at the other, and with a range of Unix workstations in between. Accordingly, the decision was taken to invest effort in a locally developed GKS, called P-GKS, which was written in C and intended to be both compact and portable. Having access to P-GKS meant that we could (relatively easily in most circumstances) mount a GKS library on any system deemed to be of interest to the University, subject, of course, to the availability of a C compiler. It should be stressed that the decision to nominate GKS as the main supported graphics library did not imply that every graphics programming application should of necessity use GKS; rather that GKS should be considered as a first choice. If GKS proved to be unsuitable for any particular application, then some other graphics library had to be considered. Also, various other libraries were maintained for a variety of purposes, for example: CALCOMP, to cater mainly for imported packages which rely on that particular interface, NAG graphics for appropriate applications, UNIRAS for users requiring 3D or high-level facilities such as axis generation, etc. We also look with interest at PRIGS and are currently considering the role that this library is likely to play within the University environment.
The issue of window systems presented a problem concerning GKS support. GKS has a static concept of the relationship between virtual NDC space and the real device space of a GKS workstation, in marked contrast to the dynamic relationships which typically exist in X and other window systems. The problems have been well documented in the ISO/IEC JTC1/SC24/WG1 Report of the Study Group on "Impact of Windowing on Graphics Standards", and we shall not dwell on them here. The problem that we had to address was how to provide GKS facilities to our users running applications under a window system? The most important area to be tackled initially was X-Windows. X was rapidly becoming a de facto standard and was being progressed along a route to international standardisation. It was being increasingly used in the EUCS environment and was viewed as a unifying interface for distributed work. Requiring users to learn the technical details of X in order to control interactions and generate graphical output was not considered to be a sensible option. Firstly, learning about the internals of X is a substantial task which could not be lightly undertaken; also, if someone did invest the effort to master X details, why should they not just make direct use of X graphical facilities and forget about GKS? For some applications, the latter approach might be the most appropriate; but the majority of users, we believed, would not wish to, or should be required to, get involved with X at a detailed technical level. Secondly, if the user's application had to run in other environments (including other widowing environments), then it would clearly be of benefit to the user if easy portability between these environments could be achieved. We decided that the only pragmatic approach was to treat X as a GKS workstation and implement a corresponding X workstation handler. This would allow GKS applications to be developed and run by users without the need to have any technical knowledge of X, other than how to use the X interface; would also allow the potential for G KS applications to run on different platforms; and would enable errors arising within the handler to be reported via the normal GKS error reporting procedures.
The next section summarises the general principles that were adopted for the construction of a P-GKS workstation handler for X-Windows.
The overall guiding philosophy that was applied to the design of the GKS X-driver was that the application programmer need have no knowledge of X-windows, apart from that necessary to operate the system. This has the advantage that the programmer will be able to make use of the facilities offered by X without having to have knowledge of the details of X; it will also ensure portability from other non-X devices. This was achieved by a single driver, which acts as a client for the X-server on the device, and covers all possible X hardware configurations by interrogating the server for the necessary information. The driver is able to handle normal X events such as resizing, exposure etc. The current handler only supports a single X window, but it is planned to extend it to allow multiple windows, each being treated as a separate GKS workstation.
X, being pixel based, creates problems of portability for applications moving from one sort of X-served hardware to another. To overcome this problem the X window on the output device was defined to be a virtual surface area of size 4096 by 4096 units (the 4096 is arbitrary; it seemed a reasonable choice, but can be easily changed). Once a physical window on the device itself is opened, a scaling factor is used to map this virtual surface onto the real output area. The transformation is performed such that the largest possible area is fitted to the output viewport without affecting the aspect ratio. If resizing of the X window takes place while the surface area is not clear, then the new window boundary will be drawn immediately, but the scaling factor is not altered, and resizing of the GKS image is delayed, until the next clearing by GKS of the window. This may have the temporary effect of either clipping some of the output if the window is reduced or leaving part of the X window empty if it is increased.
The figure below illustrates the processes involved.
When an X-window is created the server is queried as to the colour facilities available and the information returned by the server is used to update appropriate internal GKS status tables.
On any device, P-GKS constructs line types, pattern fills etc using the minimum width line, for that device, together with basic pixel operations. The P-GKS X handler is no exception, and, consequently, there is no need to interrogate the X server to determine available output attributes. We did experiment with using the X routines to set lines of various widths as provided on certain X servers, but found that the minimum line-width/ pixel technique was generally superior.
P-GKS maintains its own internal definition of its fonts and does not rely on externally provided fonts, either hardware or software based, so that the use of X had no effect on its normal method for handling text.
An internal status table is maintained just as for other P-GKS drivers, and inquiries are answered by examining this table. If inquiry requests are received prior to opening the X workstation handler, a temporary display is opened (but not mapped) and the server queried appropriately.
In P-GKS, all segments, whether WDSS or WISS based, are held in client memory and there is no need to take any special action in the X handler.
Apart from error conditions, it was found that, in practice, for level 2b GKS, only two events were of general importance: the re-sizing and the exposure events. Re-sizing events are only looked for when a window is cleared; if a re-sizing event is reported, the scaling factor transformation is adjusted to ensure that the virtual window maps correctly onto the new physical X window. When the G KS X workstation is opened, a decision is taken as to whether restoration of damaged windows is to be handled by the window system or the client (ie the X workstation handler in this case). If it is the window system, no action need be taken. If it is the X workstation handler, then the method that we have adopted is to initiate a subsidiary process which maintains the window's integrity.
Our experience to date with the X handler approach has been very positive. If we had not developed it, there would have been a significant gap in our support for GKS. Users developing GKS applications, or running imported packages utilising GKS, would not have been able to do so under X. Above all, the ability to run a GKS application in a variety of disparate hardware/software platforms is extremely valuable in a distributed environment. It is difficult to generalise about how other GKSs might implement an X handler, since each GKS implementation has its own internally defined means whereby it creates the standard GKS workstation interface. Based on our experience, however, and taking X facilities into account, we believe that it should be relatively straightforward to develop an X workstation handler for most GKSs, subject to caveats about possible compiler cross-calling problems that might arise in any specific environment. For the future, we are now looking at the development of an MS- Windows 3 workstation handler.
Two books are on track for publication in October/November.
The first is a comprehensive technical summary of the state of the art in the field of scientific visualization and is entitled
Scientific Visualization Techniques and Applications Eds K W Brodlie, L A Carpenter, R A Earnshaw, J R Gallop, R J Hubbold, A M Mumford, C D Osland, P Quarendon Springer-Verlag, 300 pp.
It is intended as a reference volume and contains chapters on Framework, Visualization Techniques, Data Facilities, Human Computer Interface, Applications, and Products. An Introduction gives an overview of the field, and a final chapter summarises the conclusions. There is also a Glossary of Terms, a comprehensive Bibliography, and a summary of Enabling Technologies.
A companion volume is entitled,
Introductory Guide to Scientific Visualization R A Earnshaw and N Wiseman Springer-Verlag, pp100.
This volume is intended for readers new to the field and who require a quick and easy-to-read summary of what scientific visualization is and what it can do. Written in a popular and journalistic style with many illustrations it will enable readers to appreciate the benefits of scientific visualization and how current tools can be exploited in many application areas. This volume is indispensable for scientists and research workers who have never used computer graphics or other visual tools before, and who wish to find out the benefits and advantages of the new approaches.
Both volumes have arisen out of initial work done at a Workshop on Scientific Visualization held earlier this year in the UK, organised by AGOCG, and reported in earlier issues of the Newsletter. Some copies of the latter volume, Introductory Guide to Scientific Visualization, are to be made available to the community via AGOCG once it is published.
How to provide graphics support in an environment where more and more diverse software products are operating across increasingly distributed systems is a problem facing many Computer Services today. It was suggested at the last AGOCG meeting that there would be substantial benefits to the community if there was a forum for discussing the issues, identifying the problems and coordinating efforts to overcome them.
It is hoped to hold a workshop later this year on this subject and the following topics have already been proposed for consideration:
I would welcome comments on the proposal for a workshop to be held later this year, suggestions for any other topics you feel are important and indications of whether you would be available to participate.
This month has an article on one of our national services: the national microform service at the University of London Computer Centre (ULCC). This article gives a very full account of the service and we are grateful to John Gilbert for taking the time to cover the subject in such depth. This article is complemented by one by harry Robertson giving a user's view of the service.
This section is intended to give space to those working in, or having a relation to, education and research to write on topical subjects. This is a good opportunity which will perhaps help you to attract support and finance Contributions should be sent to me
The national micro-form service based upon its Dicomed film recorder. This service provides high resolution graphic and alphanumeric output onto various film formats.
Essentially the Dicomed is a very high precision colour film recorder. The user may choose to receive 35mm or 16mm images in either black and white or colour. In either case, the graphics may be high resolution vector (with 32K addressability) or high resolution raster (up to 4K × 4K). Black and white supports continuous grey scale, while full 24-bit colour (>16 million) is possible. The user may, therefore, produce images running the gamut from publication quality line drawing through to photo-realistic slides. The 16mm capability is ideally suited to animations needing to exploit the high resolution capabilities of the system.
The film recorder may also be used for the production of microfiche, and the user may select 24x, 42x, or 48x reductions. Full formatting facilities (titles, indexing, pagination, etc.) may be simply specified by the user. In a typical year, the Dicomed service is used to generate in excess of 50,000 colour frames, 100,000 black & white images, and 2,000,000 microfiche pages.
To better appreciate the versatility of this system it is necessary to have a more detailed understanding of its capabilities - these are covered next.
The Dicomed D48C graphic COM system is a high precision colour film recorder. In Dl48C configuration it includes a Digital Equipment Corporation controlling minicomputer. The name D48C refers to the graphic recorder, comprising firmware controller, precision CRT, film transports and optical assemblies. The D48C is capable of image recording in vector, raster and alphanumeric modes, and may be used for either colour or monochrome recording dependent on film/optics used. It is ideally suited to many applications, including scientific plotting, business graphics (vector or raster), animation, engineering and design, and alphanumeric COM (Computer Output Microform, eg microfiche output - which may incorporate other graphics).
As a film recorder the Dicomed D48C captures the image of a variable intensity spot as it traverses a precision cathode ray tube. This CRT uses a P48 phosphor (wide spectral response), and colour is achieved by insertion of coloured filters in the optical path. For each colour filter the corresponding components of the image are traced that is, recording is in an additive manner (filters being used singly, not in combination). The filter assembly comprises seven colours (red, green, blue, yellow, magenta, cyan, and neutral) and is used in conjunction with specially designed colour corrected optics. (At ULCC colour output is currently produced by accessing the red, green, and blue filters alone and streaming the plot data accordingly.) For use with black and white film the neutral filter may be utilised; however, for the highest resolution in black and white on the aperture format a separate narrow spectral response lens is used (without filter assembly).
Three optical assemblies are in use at ULCC for graphic COM; these have been designed with appropriate specification for each of 16mm cine (colour and black and white), 35mm slide (colour), and 35mm aperture card format (black and white). (It should be noted that the term "aperture card" is used to designate an exposed image size of 37.41 × 27.94mm, on unperforated film stock. This is suitable for mounting in standard aperture cards.)
The film transport is mounted above the optical assemblies, and is able to handle 16mm (perforated) and 35mm (both perforated and unperforated) with 400 feet film magazines. This is accomplished through changes to the transport mechanics and use of appropriate aperture plates.
For microfiche applications a separate film transport is used. This primarily uses 105mm film (although it can also support 35mm roll film), and in conjunction with a dedicated electronic control unit images by column and row to form 6" x 4" microfiche (cut after processing). Separate lenses and format control are used to generate microfiche at 24x, 42x and 48x reduction. (It is possible to put six 35mm format images on a microfiche, as two rows of three. Given demand and available effort this additional service will be offered.)
In vector mode, the D48C has an addressability of 32,768 x 32,768 upon the CRT (the optics/film format will result in a maximised subset of this being available for each film format, determined by the aspect ratio of the corresponding image size). The D48C is capable of very fast vector plotting: a typical frame is captured in seconds. The time required being roughly proportional to the vector length. Exposure may be at any of 256 levels (for anyone filter). In drawing a vector, the D48C deflection system is used to accurately position the beam at selected points on the 32K × 32K matrix when the exposure logic turns the beam on for the requisite time. The vector drawn will be perceived as a continuous and uniform intensity line.
In raster mode, the D48C is able to plot on a matrix of 4096 × 4096 addressable points (pixels - ie picture elements) on the CRT (again, a subset is available to each film format) at any of 256 exposure values. The difference between vector and raster addressability is due to the spot size generated; the default raster spacing ensures a continuous image through the touching of adjacent points. The Dicomed provides the user with the means to construct display pixels as composite points with control over point and element spacing. This enables other raster resolutions to be imaged. A maximum resolution (4K x 4K) raster image in full colour (three passes) can be plotted in around three minutes.
In point mode, the D48C may be positioned to any of the CRT's 32,768 × 32,768 addressable points, which may then be exposed at any of 256 levels.
In character mode, the D48C has a high speed hardware character generator. This is particularly relevant to alphanumeric COM applications. The character fonts are stored in local memory, and depending on complexity a number of fonts may be accommodated simultaneously, with very swift font changes. Characters may be generated at different sizes and in four orientations, as standard. Speed depends on font complexity, but for typical alphanumeric COM applications speeds in excess of 30,000 characters per second may be achieved. (In practice, throughput is dependant on input rates to the D48C.) The hardware character generator functions at only one exposure level and differs from other drawing modes through operating as a stroke generator. Each character is drawn at high-speed as a series of short vectors and the CRT beam is unblanked throughout the duration of each stroke.
Intensity/exposure control is over an 8-bit (i.e. 256 level) range. However, this range is selected according to the plot mode in use: vector, raster, point, and character modes all access different CRT intensity scales in order to achieve compatible film exposure when modes are intermixed on a frame. It should be noted that hardware characters may only be generated at a single exposure level. The D48C incorporates further internal exposure tables that are selectable according to the optics and emulsion characteristics to ensure consistent film exposure under the different operating conditions. Separate table references are made according to the selected filter. These tables are programmable so that alternative emulsions may be used and the system calibrated accordingly. Indeed, calibration is required for each film type and for each optical assembly with which it may be used: this calibration is for each plotting mode and (for colour assemblies) each filter. In vector mode the internal intensity value is dynamically compensated according to the angle at which a vector is drawn. The exposure technique is time modulation which controls the density of individual points directl y through the duration for which the CRT is excited. This ensures precise control and a high degree of repeatability with consistent line width.
The geometrics of the D48C offer a very high precision with minimal distortion and a high degree of repeatability. With regard to the major axis, trapezoiding, rectangularity, linearity, and line curvature (or pin cushion distortion) are all maintained to less than (plus or minus) 0.1 %, while orthogonality of the horizontal and vertical axes is within (plus or minus) 0.15%. The spatial repeatability is also maintained to a close tolerance. Of equal importance, exposure uniformity is also constrained to a very small deviation.
The film resolution is a function of optics and film type. ULCC monitors the system performance on a regular basis. For the optics in use at ULCC the target resolutions are:
| Optics | Resolution (lpm) |
|---|---|
| 16mm | 90 |
| 35mm slide | 40 |
| 35mm aperture | 55 |
(Measurements are in line pairs per millimetre and are referenced to black and white film.)
It is important to appreciate that these are the resolutions that may be achieved at the film plane on high resolution film. It is possible to resolve 3000 lines across the image area on the CRT of the D48C. The 4 K raster of the D48C is designed so adjacent lines merge and a continuous image is perceived.
For animation purposes, the registration accuracy is held to a close tolerance, adequate for producing steady images.
In practice, measurements on the D48C at ULCC (for geometrics, photometrics resolution and registration) have consistently indicated performance well within these limits.
Having covered the functionality of the equipment, it is worth considering other aspects of the service at ULCC.
The microform operators schedule work according to system queues for each of the formats. They mount the requisite optics/transport, calibrate the system and make routine diagnostic and quality assurance checks before each run. Black and white film (including microfiche) is processed in-house; colour film goes by courier to a commercial laboratory (selected largely on the basis of their own quality assurance system). The processed film is then packaged and despatched to the end user. Black and white film is often ready the same day, while for colour film a next day despatch is more usual. However, demand upon the service may affect this (obviously, long animation runs may be deferred to overnight operation of the film recorder).
Much of the work is produced on the mainframes at ULCC - our graphics software has been designed to interface with the Dicomed and take optimum advantage of its capabilities. This is particularly important with raster output, where users are free to work in any "raster resolution" to suit their data - the Dicomed raster elements are manipulated by the system to best image that data.
With the advent of UNIRAS, ULCC has developed a generic UNIRAS-Dicomed driver. This is intended for any 32-bit host computer. A VAX specific version has also been produced. This driver is available to any academic site upon request. It requires minimal adaptation (primarily to incorporate local user identification) and has been successfully mounted at sites such as Cranfield and NERC. With the changing situation at ULCC (ie the move to Convex equipment) it has not yet been possible for our system queues to be accessed over the network. However, once the Convex service is stable, it is intended to support such access - a new version of the UNIRAS-Dicomed driver will then be made available.
There has also been user demand for simplified access to the Dicomed capabilities for imaging raw raster data. Some preliminary work has been performed for a group in Cambridge. It is hoped that effort will be available to develop a more generalised method of access for such use. Obviously, such work is dictated by demand - so if other user groups have such a requirement ULCC would be keen to hear from them.
Microfiche access is also available nationally. The Dicomed system offers a comprehensive range of formatting facilities. These enable user control of titling and indexing, both of which may be based upon information extracted from the alphanumeric data according to various positional and template criteria. Moreover, layout may be managed through fiche and column breaks - again determined by the text content. To simplify access to these facilities (and avoid dependence on the Dicomed native commands), ULCC has developed a "universal microfiche command" set. This makes intelligent use of defaults and simplifies users specification of their formatting requirements. Predefined formats are also offered; these are suited to output from particular compilers and give sensible indexes for such source data. Microfiche offers a sensible and useful way to manage/archive large data sets.
The versatility of the Dicomed equipment is such that the uses and application made of it are similarly diverse. All applications, however, enjoy output of very high quality - be they black and white (mainly vector), colour (both raster and vector), or microfiche (at various reductions).
The ease of access to the system means few users encounter problems which are brought to our attention. Consequently, we are probably unaware of many applications.
Clearly, much of the vector work is related to result presentation in graph form. Geographic information is also prominent - maps in many various forms are seen to be plotted. Output from packages such as GIMMS and SAS is much in evidence. Chemical and crystallographic structures are further common examples. Particularly heavy use is made by some meteorologists - archiving weather maps, etc. However, we occasionally become aware of unusual applications - such as printed circuit design and optical interference patterns.
Raster imaging can be even more diverse. The whole spectrum of UNIRAS output can be seen. Besides that, we see various applications pertaining to medical imaging and molecular structures. Occasionally more artistic endeavours are seen.
Animation is a demanding use of the Dicomed. We have seen many computer generated animations of such phenomena as storm development, tooth and bone growth, and other research topics, together with the production of teaching material. The system has also been used for in excess of 100 animated sequences for incorporation in Open University programmes - in areas such as Mathematics, Science and Technology, and IT. The quality of the Dicomed output has been eminently suited to broadcasting.
Use of microfiche has not been restricted to large program listings and output data sets. Many disciplines have taken advantage of the medium having particular regard to the economy of reproduction, handling and postage of large data sets when recorded on film in this way. We are aware of many scientific users, but also from other disciplines such as archaeology and anglo-saxon studies! As with all formats, the imagination of the user is the limiting factor.
At ULCC, we take pride that we are willing to help all users to take advantage of the Dicomed in a manner best suited to their own work. We are always willing to discuss and advise and, where the need arise, we will help the user develop their application. (It has to be added, manpower permitting!)
At present, it is ULCC policy not to charge for academic users for film produced on the Dicomed. However, externally supported users may be liable to charge, while the right is reserved to levy charges for abnormally heavy usage. Where the user's institution makes a charge for the service, ULCC may impose a charge on that institution.
ULCC intention is to continue to make the Dicomed facilities available as widely as possible. To that end we are always receptive to suggestions for developments and enhancements to the service, and to ideas for further expansion.
The Dicomed camera system at the ULCC is capable of producing 35 mm slides from UNIRAS. A number of slide making devices are available on the market, but the Dicomed is a high-end system with a resolution of 4018 by 2712 pixels supporting 24-bit colour from UNIRAS. The slides are of exceptional quality and appear very sharp, even when projected onto a large screen.
The original Dicomed D-driver from UNIRAS was supplied to CHEST sites with Version 5.4, but it was found that a number of enhancements and changes were needed to make the driver compatible with the customized DIMFILM system at ULCC. This work was carried out by John Gilbert and colleagues at ULCC and the driver tested by NERC, Cranfield, Oxford and others. The ULCC Dicomed D-driver runs unmodified under UNIRAS Version 6, but is limited to the V AXNMS platform at present. Evidently, there are plans to port the driver to Convex.
Since the Dicomed is being driven as a true raster device, we find that the UNIRAS plot files can be rather large. For example, depending on how the UNIRAS software is used, we can expect a raster database (RDB) file up to 30 Mbytes and, from running the Ddriver, a Dicomed plot file up to about 3 Mbytes.
At present, the plot files are sent to ULCC on magnetic tape and the slides returned within about five working days. This will be streamlined in future when plot files can be transferred via JANET, but development of the link is inhibited in the short term due to financial constraints. The Dicomed service is an important graphics facility in the academic community and there is considerable scope for expanding its usage.
After fifteen years in Computer Graphics, this was my first visit to SIGGRAPH. No one can fail to be impressed at the sheer scale of the event - several thousand attending lecture sessions, several hundred attending each tutorial. Even so, the conference was upstaged by its setting - Las Vegas - where hotels have 3000 rooms and 4000 slot machines and the casinos take in 200 million dollars a month. SIGGRAPH made just a perturbation on its brassy exterior. The price of normal life in Las Vegas is cheap - a good dinner can be had for less than 10 dollars - because meals and hotel rooms are just loss leaders for the attached casinos.
The conference itself is a busy event. There is the papers session, the panels session, the vendor's exhibition, plus this year a special track for educators and an exhibition on Tomorrow's Realities, covering virtual reality and hypermedia. All run in parallel, together with countless meetings on special topics. Experienced SIGGRAPH attendees carry a large Filofax to tell them where they should be at any time. This report, therefore, can only be a personal impression of the cross-section I encountered.
The papers session is the technical focus. There is just a single stream and hence relatively few papers considering the size of the conference. I enjoyed the session on Illumination and Reflection, where the quest for ever greater realism continued. There were two excellent papers from Cornell (He et al, Sillion et al) which are worth studying in the proceedings. He et al describe a new reflection model applicable to rough surfaces; an interesting feature is the attempt to verify the model against experimental measurement. Also in this session, Hanrahan described a hierarchical form of the radiosity algorithm. This makes use of new work on the classical N-body problem where the order N*N calculations have been reduced to order N. Hanrahan applies the same ideas to the radiosity problem, to reduce the number of form factors that have to be determined.
Another interesting session was that on Volume Sculpting and Rendering. A paper by Galyan and Hughes (co-author of the new Foley, van Dam, Feiner and Hughes textbook) showed how the metaphors of painting programs can be extended to 3D - for brush read sculpting tool. Two papers on volume rendering followed - by Wilhelms and by Laur - both seeking to improve the rendering speed of current algorithms. The favoured approach for volume rendering has shifted from the Levoy ray casting method, to the projection type methods where voxels are projected into the image plane. Laur's paper described an extension of the splatting technique introduced by Westover at SIGGRAPH 90.
There was a large number of panel sessions - perhaps more than there was material for. There was one interesting session on Future Directions of Visualization Software Environments, in which leading figures associated with AVS, apE, Khoros and IRIS Explorer compared notes on the good and bad points of their respective systems. Some interesting points were made, but few of the theatrical exchanges one might have expected from competitors - the gloves stayed on! Here is an area of concern for the user however - there are four of these similar systems (five now that IBM have announced a product). This is reminiscent of the situation in the mid-1970s with competing graphics packages such as GINO-F and GHOST - a situation that led to the need for graphics standards. It may be time to look at standardising these visualization systems.
The vendor's exhibition was certainly theatrical, with leading companies reputedly employing professional actors to communicate their message. HDTV was in evidence, and Sony showed a monitor with 2K by 2K resolution - cost about $40K. Silicon Graphics launched their Indigo workstation - which makes good performance graphics affordable (UK cost around £8K). There was a PEX interoperability demonstration: several vendors took part in this demo where clients (typically AVS) ran on one machine and the server on another. Tektronix displayed a new PEX terminal.
The Virtual Reality and Hypermedia exhibition had many interesting items. There was a collection of exhibits from University of North Carolina at Chapel Hill, including a mountain bike with force feedback that people could use as indoor exercise - but gaze at beautiful scenery via their head mounted display. (Rendering was done using the Pixel-planes 5 machine - giving 2 million polygons per second, probably the fastest rendering machine?) Another exhibit allowed one to hold a conversation (on video) with a sultry French girl - who became even more sultry if you chose the right words (or so I was told).
The educator's track ran the full length of the conference, with special panels and papers targeted at those who teach computer graphics. Outstanding for me was the paper by Donna Cox from NCSA who showed the two animations created by her class of arts students - each produced in one year starting from scratch and each of the highest quality. Not only did she show the finished product, but also some of the early efforts that were discarded or improved.
The educator's papers are published in Computer Graphics Vol 25 Number 3, and the proceedings of the conference itself in Number 4.
PS In case you are wondering, I am pleased to report that after a serious initial setback, I left Las Vegas with a small profit of 50 cents on the week.
This year's SIGGRAPH took place in Las Vegas, Nevada, gambling capital of the USA. A somewhat unsuitable venue but nevertheless the conference had its high points. The paper sessions were mostly of their normal high quality with a few outstanding ideas. The short paper presentations (20 mins) introduced last year were continued. This was generally felt to be a good idea as more papers were presented albeit without some of the mathematical details. There were a notable lack of natural phenomena papers (have we simulated everything?) and Radiosity papers are still abundant but getting more mature with a trend to towards the application areas.
Cornell University graphics researchers under Don Greenberg teamed up with physics researchers to give us a better physical model for light reflectance (specular and diffuse). The most obvious result was that specular images can be shown to decay with surface roughness. The model was also incorporated into a radiosity method. There were some good animation papers with David Baraff completing his hat-trick of SIGGRAPH appearances adding friction to his model and producing some realistic collisions between block objects. A paper from MIT Media Lab showed how a filtering technique can produce apparent motion in objects without actually moving them. This has some interesting applications including air traffic control displays which could show the direction of travel of an aircraft without changing the position of the icon or cluttering the display with arrow indicators.
An excellent idea from NIT was to produce an NC machine path from Zbuffer information. This method was effectively demonstrated as the authors (Saito and Takahashi) were able to produce actual physical models including the famous Newell teapot. A limited edition of 100 model teapots were distributed at the conference to the chosen few! Also from Japan, Muraki of the Electrotechnical Laboratory described a method for using blobby objects to make a model from range data. He showed some very effective faces from scanned data. On the subject of implicit surfaces, Jules Bloomenthal and Ken Shoemake from Xerox P ARC came up with a good idea they call: convolution surfaces. These surfaces are seamless and allow complicated models to be built from skeletons. Blobby or soft objects were also popular with a team from Brown University who introduced an interactive sculpting technique using a 3D (Polhemus) brush to trace out positive and negative paths in volume data.
Perhaps the most agreeable visual effects were produced by two independent groups who generated textures using the Reaction-Diffusion technique. Greg Turk from UNC and Witkin and Kass from Carnegie Mellon and Apple, made use of a technique proposed by Alan Turing to account for pattern formation in biological morphogenesis. The idea is that two or more chemicals diffuse at unequal rates over a surface and react with each other. Some startling effects were produced by both groups including the synthesis of finger prints! A third texturing paper by van Wijk from ECN in the Netherlands, gave us another technique, "Spot Noise". This turns out to be excellent for visualising scalar and vector fields which was amply demonstrated by van Wijk on video. In the "hands and legs" session Dean Rubine from CMU attracted much audience appreciation when he showed how his system could learn complicated gestures from the user and apply these to applications such as composing musical scores. Another paper from the Netherlands (SCAN) by Rijpkema and Girard described some good work on the animation of human grasping and the roboticists from MIT (Raibert and Hodgins) put their legged creatures through their paces in simulation and as physical robots. Cary Phillips gave the last paper (with Norman Badler from U. Pennsylvania) and showed some interesting interactive control mechanisms for a human figure. He also observed that this was his third SIGGRAPH paper and the third time he had been the very last speaker of the week! (SIGGRAPH organizers please note!)
I did not attend any of the panel sessions but I heard various reports that the intellectual property rights session was good and raised some very worrying issues in the light of various law suits and patents of hitherto un-patentable algorithms. Another hot topic was the merging of HDTV with computer graphics.
The electronic theatre had some of the best animation I have seen but also some of the most mediocre. I didn't see much in the way of totally novel motion control techniques, although the scenes from Terminator Two were truly awesome! There was a lot of wonderfully detailed models, particularly the dinosaurs and others in Lost Animals and IGT's scary dragon. The Invisible Man in Blind Love from Georges (Eurocitel) was done in beautiful black and white mystery movie style and avoided the human body animation by making the main character invisible. They also used a viewing algorithm that didn't commit the VUP vector to the Y-axis so there were some interesting camera angles in keeping with the genre that was being imitated. Pixar and John Lasseter were there again with a wonderful demonstration of his mastery over the media with a revisit to our friends Luxo and Junior in Luxo Jr. in Light & Heavy and Surprise. There were a number of good scientific visualizations, particularly Evolution of Gravity and Effective Topography on Phobos by Wayne Lyttle from Cornell. The show seemed too long this year and had a high content of arty pieces some. of which were long and boring and did not demonstrate anything new. In general I enjoy the art pieces in the electronic theatre, but the balance was wrong this year. An excerpt from the Screening Room was shown and contained some good pieces, definitely something the organisers should keep in next year. Also please put the 3 D stereo pieces together so we don't have to keep putting the glasses on and off.
At the start and interval of the electronic theatre, Loren Carpenter came up with a wonderful audience participation idea. Each member of the audience was issued with a paddle containing red and green reflectors which could be picked up by an online vision system and used as input to an SGI workstation. Games of Pong were possible with thousands of participants on each side. Less successful was the attempt to land a jet in the flight simulator. Not surprising with no audience control over throttle, spoilers and flaps! This was a wonderful idea although for most people it went on a little too long.
Beau Takahara of George Coates Performance Works presented a multi media event in which the stage was viewed in stereo with suitable glasses. The piece entitled Invisible Site had contributions from many people and had some good ideas. Most of the audience members that I talked to enjoyed the first 20 minutes or so but after that there just didn't seem to be enough new ideas to justify an hour performance. The performers and musicians were excellent but the use of standard SGI demos with vague interactions with the performance could not sustain the pace. I would encourage this type of show but it seemed that there was a lot more that could be done in getting the actors to interact with the 3D images.
Two new events took place: an Educators Program, which I did not attend (I heard some good reports) and an exhibition entitled Tomorrow's Realities. The latter had some terrific demonstrations with interaction ranging from a ride on a surf board through the more familiar data gloves to a ride on a bicycle! An innovative and exciting exhibit.
Finally the trade show held my attention for a day and there were some good new products. HP introduced their SNAKES 72 MIPS workstation (awesome power for a reasonable price) and SGI introduced the Indigo. IBM made a good showing this year with their visualization demonstrations. Some good animation software was demonstrated as usual from Alias, Wavefront and SoftImage and some new software from Acrobat of London attracted a lot of interest. I was there to buy a film recorder and the top of the line machines were outside my budget. 2Film Technologies of Ontario, Canada had a machine for an incredible $3k which was 24 bits and 2k resolution. There was nothing else to touch it. Video animators still await a cheaper version of the Abekas and the video disk technologies don't quite seem to be there yet.
To sum up I thought that there was a noticeably variable quality in both the Electronic theatre and the technical papers, with fewer people in attendance at the technical sessions than in earlier years. Perhaps there was just too much happening or the lure of the gaming tables was too great. A croupier was overheard to say " ... don't know why these SIGGRAPH people came here, most of them don't gamble and those that do count cards!".
