3D --- Media Creation --- Shared Space
Published by 4th WAVE, Inc.
Issue #709 4/25/97
CONTENTS
****************
To Our Readers
Some of you may have gotten multiple copies of the last issue. Apparently our listserver provider had a software problem when the issue was sent out. We apologize for any inconvenience this may have caused.
****************
**LATE NEWS**
The WAVE Report has received word that ZD-BOP will be releasing 3D WinBench 97, their 3D benchmarking application that is the successor to MonaLisa, today at 5pm. This software is expected to be very important in 3D benchmarking for the industry. We will have a report on it in the next issue. It is expected to be available for download today after 5pm from their Web site at:
http://www.zdnet.com/zdbop/3dwinbench/3dwinbench.html
*************
Real3D To Support Sense8 [April 18]
Real3D and Sense8 Corp. have announced that Real3D will support Sense8's WorldToolKit software on their PRO-1000 Series of 3D accelerators. Software deliveries are expected to coincide with this year's SIGGRAPH 97 in LA (early August).
Diamond Ships Monster Sound [April 21]
Diamond Multimedia Systems, Inc. have begun shipping the Monster Sound PCI-based add-in audio card. The card, which will sell for $199.95, utilizes A3D Interactive audio from Aureal Semiconductor to offer multi-stream positional 3D audio.
Monster Sound was designed specifically as a DirectSound and DirectSound 3D accelerator. According to Diamond, the card offers quad- speaker output that creates sound similar to home theater surround sound systems. It should be available at most retail stores this week or the beginning of next week and will be bundled with three complete game versions of Lucas Arts Entertainment's Outlaws, Maxis' SimCopter, and GT Interactive's Tiger Shark. For more information on Monster Sound, check out Diamond's website or watch for a review on MonsterSound in a future issue of WAVE.
Microstation 95 for Macintosh [April 7]
Bentley has released the successor to Microstation V5 on the Macintosh: Microstation 95 for Macintosh. The new version incorporates the functionality of the Windows 95 version as well as additional capabilities, making it the most complete release of the product to date. New features include a streamlined graphical user interface, the new AccuDraw and SmartLine drawing technologies, and a multi-platform MicroStation BASIC language and development environment. MicroStation 95 for Macintosh also allows users to attach binary and continuous tone raster files and perform basic warping functions; plotting and printing are improved through the support of pen-tables, configuration files and plot priority settings; and its 3D design and modeling functionality now includes key frame animation and the improved performance of hidden-line removal.
3Dfx has filed for an IPO 4.2 million common shares with the SEC, and has announced that Sega will purchase 700,000 of those shares. Earlier in March it was announced that 3Dfx will design the 3D technology to be used in Sega's next home game console. Shares are expected to come to market at around $8.50.
Intergraph Brings Game Development Software to NT [April 22]
Intergraph has announced an OEM agreement and partnership with Nichimen Graphics to bring Nichimen's N-World game authoring software to the NT platform (previously it was Unix-only). In addition, Nichimen has licensed Intergraph's RenderGL technology to incorporate in future versions of their software. Three main components of the N-World software are being ported to NT: N-Geometry, N-Paint 3D and several exporting tools. The NT package will be marketed under the name "Fast Track." Availability has not been disclosed.
Cambridge Animation Systems - Animo Ax-Cel Announced
We missed the NAB announcement of Animo Ax-Cel by Cambridge Animation. This new software comes as a result of its working relationship with Warner Bros. Feature Animation and DreamWorks SKG. A completely new program written for the SGI O2 workstation, this software is being used for feature film animation production. A central feature of the Animo approach is the exposure sheet (Xsheet) and layouts. Animo Ax-Cel adds new capabilities which include Digital Compositing, Multi-Tasking, improved speed with the ink and paint tools, and support for the International Color Consortium (ICC) color tagging. Animo Ax-Cel has a floating license capability which allows it to move to any machine. The software will be available in May with the core module costing $12,000. A rendering and image processing server costs $4,500 and plug-ins cost from $1,500 to $4,500.
e-mail: admin@cam-ani-us.com
Intervista Teams With Bitstream [April 22]
Intervista Software has announced a strategic partnership with Bitstream Inc. in order to "improve the quality of text in 3D Internet applications." Intervista has also licensed Bitstream's TrueDoc font compression and portability technology for inclusion in their future products. In addition, Bitstream and Intervista have agreed to work together to promote Bitstream's portable font resource specification for standardization within the VRML consortium.
Vertigo Uses Thinkfish Renderer in Photoshop Plug-in [April 22]
ThinkFish Productions has announced that it has licensed its LiveStyles 3D artistic rendering engine to Vertigo Technology for use in Vertigo's new Photoshop plug-in. LiveStyles will be included as a new rendering option within Vertigo 3D Dizzy version 1.0, a plug-in for putting 3D models into Adobe PhotoShop. Vertigo 3D Dizzy 1.0 is available immediately at a retail price of $99.95 and will ship with five ThinkFish styles free of charge. Additional styles will be available from the ThinkFish Web site for $24.95 per style or $59.95 for a full family of styles (includes 5-7 styles).
Spacetec IMC Ships Animotion 2.0 [April 7]
Spacetec IMC Corp. has released SpaceWare AniMotion 2.0, the next- generation version of their 3D interactive motion control plug-in/tool for 3D Studio MAX. AniMotion 2.0 is a hardware/software solution for creating and recording complex animation, such as falling leaves, flying aircraft or twisting logos. New features include the ability to manipulate a camera for view control around selected objects, automatic spline creation, time-based recording and the support for new 3DS MAX 1.1 features. AniMotion 2.0, which ships with the Spaceball SpaceController 3D input device, is supported on both Windows NT and Windows 95. It is available now for $495 through Kinetix VARs.
Viewpoint Makes Models Available in DirectX File Format [April 28]
The attendees at the Computer Game Developer Conference Viewpoint will be able to get a CD-ROM with 4,000 quality 3D models and 200 animations in the DirectX file format. This is another step in implementing the license which Microsoft has with Viewpoint. Viewpoint has also been working with Microsoft to develop the specification of the DirectX format.
Beginning May 1, the 2,000 pre-licensed models may be seen and accessed at:
http://www.microsoft.com/directx
or
Viewpoint also announced DataShop for Real-Time, a CD-ROM collection of 1,000 models in Multigen .flt, Coryphaeus .dwb, and Evans & Sutherland .gdf file formats and DataShop for Developers, a CD-ROM with over 2,000 new models optimized for real-time performance in games and multimedia applications. For a limited time, developers can license DataShop for Developers for $2,495
Newfire Unveils Catalyst [April 22]
Newfire has released Catalyst, an authoring tool for creating commercial-quality, 3D games for the World Wide Web. Catalyst will debut at the CGDC next week. With the tool, game authors will be able to create games for the Internet using popular 3D authoring applications, while the underlying technology is based on VRML 2.0 and Java. The base platform will include scene composition, an interactivity editor, real-time playback and analysis, a performance optimizer, and a project manager. The base product with a single-user license is $1,995 and will be available to developers in June.
Newfire's 3D Game Contest [April 25]
Newfire has announced that they are holding a contest to find the "hottest 3D game content for the Web," called the TorchIt! game competetion. Participants can compete for various hardware and software prizes. The entry deadline is July 28. More information can be found on Newfire's Web site.
Computer Game Developers Conference
Just to remind those that may have forgotten:
The CGDC will feature 100+ exhibitors and 200 conference sessions, along with an estimated attendance of 5,500. WAVE will be attending the conference and will report on it in the next issue.
The Quake Tournament is co-sponsored by Intel and TEN. The grand prize is a 200 MHz Pentium PC with MMX technology.
International Conference on Informational Visualization-IV97
The theme for this conference will be "From Research to Practice," with planned symposia including Fundamentals of Graphics & Visualization, Visualization in Network, Computer Aided Geometrical Design, Virtual Environments, Visualization in knowledge- & Data-bases, Animation and Special Effects, Education of graphics and visualization, Visualization in Humanities, and Visualization in Construction.
Tel.: +44 171-815-7476
email: banisse@sbu.ac.uk
709.3 3D Accelerator Performance Characterization by David Lohse and John Latta
The last few years have seen a surge of interest in 3D on the PC and as a result, a corresponding level of interest in characterizing the performance of 3D chips, cards and systems. Much of this interest stems from the desire to obtain the best cost performance in 3D similar to what is done when buying computers or 2D graphics cards. However, 3D is more complex than simple single number metrics. Although we are convinced that some of the performance measures will reduce to this, other aspects of quantifying performance will not be so easy - in particular, image quality. Thus, it is important to look at 3D performance as more than just benchmarking but characterizing the performance of all that generates the image. The PC industry has only begun to address the range of issues required to both quantitatively and qualitatively assess performance.
At 4th Wave (parent company to WAVE), we have begun to examine how to characterize 3D performance. Our initial efforts started by gathering 3D benchmarks and focusing on the best. These early efforts run the gamut from testing various aspects of the 3D pipeline up to the scene- level output. We have extensively tested several 3D accelerators and created a preliminary report. In this issue we present an overview of the report and results. The full text of the report, along with our 3D benchmarking results (including comparative graphs) can be found on our Web site at:
http://www.fourthwave.com/3d-perf
Our current suite of tests is comprised of eight different 3D performance applications. They are all independently developed by industry-leading 3D hardware and software companies as well as the recognized Standard Performance Evaluation Corporation (SPEC):
Benchmark Organization Type of Organization --------- ------------ -------------------- D3Dbench 3D Labs 3D hardware vendor GLPerf SPEC Independent benchmarking organization GL Quake id/3Dfx Software developer/3D hardware vendor Realibench Datapath 3D software vendor RWG Benchmarks Gemini 3D software vendor ViewPerf SPEC Independent benchmarking organization WizMark 3Dfx 3D hardware vendor WTK Direct Sense8 3D software vendor
The 3D graphics pipeline is a complex technology, consisting of multiple stages, including the host system's processor and memory to varying degrees based on the design of the 3D accelerator. To better understand this pipeline we have taken three approaches to 3D performance testing:
* Pipeline * Scene-level * Application-level
The 3D pipeline of a given chip set normally has many different rendering options can affect the output, both in terms of performance and visual quality. Although options, such as lighting, culling, fogging, etc., are available most benchmark software tests a limited range of these. In our suite of tests we focused on the most important, in terms of visual quality, of these options, the texturing and filtering modes. The result is usually large performance penalties the more features that are turned on.
Here we are interested in the final output of the 3D pipeline. There are really two parts, the frame rate and image quality. While the frame rate is an easily-definable and measured parameter, visual quality remains subjective. Ideal visual quality observations would require side-by-side monitors running the same test from identical reference platforms, but the practicality of doing this on a large scale is low. At this stage of our testing we instead focused on obvious visual cues, such as general appearance, texturing, occlusion, fogging, etc. In future efforts we plan on incorporating a "side-by-side" testing scheme.
This level comprises the final end-user performance of a commercial 3D- based application such as CAD, modeling/animation and visualization software. It consists not only of the performance of the 3D subsystem, but is also intimately connected to the host processor, system, memory and operating system, and most importantly the application software itself.
Currently we have not included any application-level tests in our suite. In future efforts we plan to incorporate applications from the important 3D application categories, including at least CAD, modeling/animation and visualization. We note that is will be an especially difficult area because benchmarks are best done on those applications which are the most taxing in terms of 3D performance. However, today there are no such mass market applications, outside of games. Applications such as 3D Studio MAX could offer an excellent test opportunity but this is not a mainstream product. Thus, 3D acceleration is leading the market, at least at the application level of testing.
Our tests have been used to compare 3D accelerators in several different ways:
* Cross-Vendor - the performance differences between different vendor's products * Cross-OS - differences in performance under two different operating systems (Windows 95 and Windows NT) for the same accelerator. * Cross-API - differences in performance under different 3D APIs for the same accelerator.
The most obvious type of comparison that can be evaluated are the performance differences between different vendors' products. This encompasses performance using both standard metrics (polygon rate, etc.) as well as visual quality. Since the same hardware reference platform (host system), operating system and software, and benchmark software was used consistently for each test, valid comparisons between vendor products may be made. However, it is important to examine results from all 3 levels of testing as outlined above in order to obtain a fair assessment of the performance differences.
3D graphics on the PC has for the past several years been primarily the domain of Windows NT, which natively supports the OpenGL API. With the introduction of Windows 95 and the DirectX APIs, however, 3D has become more important on the consumer Windows platform.
Here we attempt to examine the differences and performance issues between tests run on the same host system under the two different operating systems (Windows NT and Windows 95). Although most tests do not currently run under both operating systems, we have analyzed the important ones that do in order to showcase these differences.
The issue of API performance is a hot topic in the industry as the leading APIs fight it out to become the primary API. While currently most attention is focused on SGI's OpenGL and Microsoft's Direct3D, other more proprietary APIs still rank a level of importance, including 3Dfx's Glide and Rendition's Speedy3D. Given that Speedy3D only runs under DOS it expected to have a minor role.
Here we attempt to compare API performance directly by running identical tests under different APIs. Currently the only benchmark test available with these capabilities is Datapath's Realibench, which supports all of the major APIs and provides both metric (frame rate) and visual quality feedback.
For all tests, the reference platform used was a 166 MHz Pentium with 64 MB RAM.
For our initial performance characterization efforts we examined the following 3D accelerator boards:
Board Chipset
3Dfx Obsidian 50-2440 Voodoo Graphics 3Dfx Obsidian 50-4220 Voodoo Graphics 3Dfx Obsidian 100-4220 Voodoo Graphics Symmetric GLyder MAX Permedia + Delta 3Dlabs Racer 500TX + Delta Canopus Total3D Rendition Verite V1000L
For full hardware specifications check our Web site.
Due to driver, API and OS limitations, most of the tests in our suite were not available for all of the boards we tested. The only two benchmarks that were available for all of the boards we tested were Realibench and D3Dbench; GLPerf and Viewperf only ran on the two 3Dlabs boards but were valuable for at least comparing these two chipsets (500TX and Permedia).
In all of the tests we ran, the Obsidian boards consistently gave the highest performance, in many cases by a wide margin. The tests were run over increasingly complex forms of texturing, from untextured up to trilinear filtered. Of the Obsidian boards, the two 50- level boards handled all forms of texturing without significant change in performance, but surprisingly the high-end Obsidian 100-4220 took a drop in performance from texturing with the Glide API (not true using Direct3D, which only supported bilinear texturing), especially when the tests moved into filtered textures.
The 3Dlabs boards, on the other hand, were affected more by texturing than the Obsidian boards, and the Permedia handled texturing much better than the 500TX-based board. The 500TX board took large performance hits from the increasingly complex texturing, while the Permedia's performance dropped only slightly. This resulted in the 500TX board giving higher performance for the untextured tests, while the Permedia board offered superior performance for filtered texturing.
Realibench was run on all of the boards on the major APIs they each supported, including primarily Glide, D3D and OpenGL. The results showed one consistent result - D3D offered lower performance on all of the boards than either Glide or OpenGL. One problem that potentially affected the results for one of the databases used was that of screen refresh. When double buffering, the switch between the buffers is locked to the refresh rate of the monitor by default, resulting in frame rates that are integer fractions of the monitor refresh. This caused a problem for one database which potentially could have resulted in a higher frame which was instead limited to 30 fps. By setting an environment variable for the Obsidian, this effect can be avoided by allowing the buffers to switch as fast as possible, but D3D does not currently offer a similar work-around and is therefore limited by these effects. According to Microsoft, there is an important caution in allowing the image update rate to run asynchronously with the frame rate - there can be distracting image tearing and image artifacts due to the switching between buffers. Microsoft claims that this is one reason while D3D does not allow such changes and should not used against it in these tests.
D3D still offered lower performance in almost every case but a few. The different texturing modes did not appear to offer any distinct performance differences between APIs - i.e., the performance differences remained constant or sporadic for each type of texturing.
Cross-OS Results and API Implications
Since Realibench is supported under both Window NT and Windows 95, it allowed us to directly compare the 3D performance between these two operating systems. For the Obsidian boards, the Glide API was used, and for the 3Dlabs-based boards OpenGL was used (since these two APIs are cross-platform).
For the Obsidian boards (for which Glide was used on both platforms), the WinNT performance was always at least slightly higher than the Win95 performance, while for the 3Dlabs-based boards (which used OpenGL on both platforms) the Win95 performance was consistently higher.
These performance differences can possibly point to any of several reasons:
* The most obvious implication to assume would be that the APIs are optimized more for one platform than the other, but since in these tests OpenGL performed better on Win95 than on WinNT, this assumption seems unlikely. * Differences in the drivers, which may be more optimized for one OS than the other. * Differences in Realibench performance itself, which uses the same code under both OSs * Differences in 3D performance itself under the two OSs, particularly in how 3D data is passed to the host processor and rendering engine.
Visual Quality and API Implications
Since side-by-side tests were not run on the boards, the visual quality cannot be effectively evaluated between them. Some of the configurations (board & API) resulted in obvious quality differences, but these appeared to be more closely related to the API & driver implementations on the boards. Specifically, on the Obsidian boards the Glide API, which is optimized for the Voodoo architecture, of course resulted in nearly flawless images.
Using D3D on these boards however, resulted in several visual problems: the Helisim database was completely washed out, which according to Datapath is due to the fact that Obsidian does not handle fogging correctly under D3D, and some bad texture mapping and object occlusion could be seen on several of the tests. In fact, D3D caused image quality problems on at least some of the tests on all of the cards, usually focused on the texturing such as mis-aligned textures or texture maps replaced with solid black or blue planes.
This test, which was Direct3D-only, was able to run on all the boards we tested since they all supported D3D. The tests were run over a range of triangle sizes from 5 pixels up to 455 pixels, and over three types of texturing (untextured, point sampled and bilinear filtered). Smooth shading was also used for all of the tests.
We split the tests up into two primary groups - meshed and unmeshed - in order to examine the impact that polygon meshing has on polygon rate performance. Not surprisingly, the tests in which meshed triangles were used (mesh size of 30 triangles) resulted in superior performance on all of the boards, particularly for smaller triangles. For triangle sizes under 100 pixels, meshed triangles offered performance as much as 50% higher than unmeshed triangles, but for triangle sizes greater than 100 pixels, the effects of meshing became almost negligible, with the meshed and unmeshed tests offering nearly the same performance, although the meshed tests were almost always at least slightly consistently higher than the unmeshed tests - the one exception to this is the Obsidian card, on which the unmeshed tests slightly edged out the meshed for triangles larger than 200 pixels.
Again, the Obsidian card outperformed all of the others in the test suite by a significant amount on all of the tests, giving unrivaled D3D performance. The Obsidian's advantage was particularly superior for the more complex texturing modes - its performance dropped only gradually for increasingly complex texturing modes, while the 3Dlabs boards took heavy performance losses once filtering was introduced.
For the untextured tests, the two 3Dlabs-based boards performed almost identically, while the Total3D (understandably) lagged slightly behind. However, once textures were introduced, the Permedia-based board pulled ahead of the 500TX, whose poor capabilities at handling textures caused it to fall behind even the Total3D once bilinear filtering was introduced.
Since GLPerf is OpenGL-based, it was only possible to run on the two 3Dlabs-based boards, however, it resulted in several interesting observations.
Texturing and Triangle Size Implications
For this test we had GLPerf calculate the polygon rate for runs over triangle sizes ranging from 5 to 400 pixels and over increasingly complex texturing modes from untextured up to trilinear filtered. As expected, the increasingly complex texturing modes resulted in increasingly lower performance results. However, the trilinear filtering tests should be ignored in any case - the 500TX does not support trilinear filtering under OpenGL, since OpenGL requires per- pixel MIP-mapping which the 500TX does not support, and the Permedia does not support trilinear filtering at all.
Confirming the effects of triangle size on performance seen in D3DBench, in GLPerf the "magic" number also seems to be around 100 pixels. For triangle sizes under 100 pixels, the performance is significantly higher than for larger triangles, a benefit which decreases quickly until sizes of around 100 pixels are reached, at which point the effects of increasingly large triangles levels off. This is interesting to note in that both Direct3D and OpenGL accelerate optimally for triangles of size less than 100 pixels.
The triangle size and the texturing mode do not appear to have any correlated effect on performance. i.e., increasingly larger triangle sizes have the same performance impact on bilinear filtered tests as they do on untextured ones.
Since GLPerf is executable under both WinNT and Win95, it provides us with interesting comparisons between performance under the two OSs. Unlike the results found in the Realibench cross-API testing, the GLPerf results are closer to what is expected - in most cases, the performance for both 3Dlabs boards was very close under the two OSs, with WinNT performance slightly edging out Win95 performance.
The exception to this was again the 500TX board. Although it performed fine for untextured polygons, with WinNT offering slightly better throughput, once texturing was introduced it suffered particularly under WinNT. For all texturing modes (point sampled and bilinear), the 500TX board's Win95 performance rose over the WinNT performance for triangle sizes larger than about 50 pixels (i.e., for moderate- to large- sized polygons). It was already noted that the 500TX handles textures poorly, and this indicates that this is an even more pronounced problem under WinNT than under Win95, most likely due to driver implementations under the two OSs.
ViewPerf was also only available for the 3Dlabs-based boards, since it relies on the OpenGL API.
The most useful interpretations that can be drawn from our ViewPerf tests on the 3Dlabs-based cards are their variation in performance between Win95 and WinNT. Again, in most cases for both boards the WinNT performance was consistently higher, as expected. Although there are a few exceptions, the tests run under WinNT are greater by less than 30%, and for both the CDRS-03 and DX-03 viewsets, the average scores are around 12% greater for both boards run under WinNT.
The DRV-04 viewset, on the other hand, presents some anomalies. For this viewset, the average scores are actually lower for WinNT than for Win95 (3% and 10% lower for the Permedia and 500TX, respectively). It is not clear what the cause for this is, although it is possible that either the OpenGL implementation on Win95 or the Win95 drivers are more optimized for handling the complex coloring used in this model.
For the full text of this report, including descriptions of the benchmarks and 3D accelerators, as well as tables and charts of the benchmark results, check out:
http://www.fourthwave.com/3d-perfom
709.4 Company Profile - RTIME - Providing the Foundation for Large Network Games by John Latta
The WAVE Report met with Chip Overstreet, President and CEO of RTIME to learn about new initiatives the company is making to bring its API set for network game play to market. The company has developed an interactive networking engine which enables scaleable game play over the Internet. Its core product is the RTIME SDK which provides an object-level interface for game developers. The architecture of RTIME is based on four components: a client/server relationship, an affinity- based data distribution in the client/server context, a dynamic motion model tailored to each client, and bandwidth management. RTIME feels that its use of a client/server architecture allows it to treat the server like a network switch for game play. Servers can be distributed both in proximity to the clients and in the virtual world which can divide the play space geographically. It is the concept of affinity, the proximity of objects in virtual space, which is implemented in templates, that allows RTIME to have a highly tailored description of clients based on location, priority, data types and even user-defined sub-types. Using its dynamic motion modeling, RTIME can allow clients with different capabilities to run asynchronously on the network. RTIME also keeps a global clock and each update is time stamped. The ability to handle dynamic motion extends to the level of objects and several means are provided which include dead reckoning and path-based motion updating. The developer can also implement specific approaches to dynamic motion modeling.
RTIME released its first product, Version 1.0, in October 1996 and Version 1.1 came out in January 1997. Clients are available for Apple MacOS 7, Windows 95, Windows NT and SGI. The server runs on Windows 95, Windows NT and SGI. Support for SUN is coming. Version 2.0 which includes UDP (User Datagram Protocol) support and usage capture is due to be released in the summer.
With Version 2.0, RTIME claims that its software is scaleable to support up to 1,000's of players on the Internet. Two licensing models are available. RTIME 64 is focused on retail free online play. It supports up to 64 simultaneous players per server. Pricing is on a box basis and not based on usage. The second is RTIME Unlimited which is geared for commercial games and includes the ability to capture information for billing. There are no restrictions on the number of players and the total cost of the license is approximately $.25/hour of game play per person. The SDK is provided free for developers.
To show the potential of RTIME the company has recently released RTIME Rocks!, a fast action game capable of supporting up to 50 players and 50 observers on the Internet. Observers and enter to play or exit at anytime. A version for Windows 95 is downloadable from its web site.
Recently RTIME closed a round of financing with Pacific Horizon Ventures and others at $1.5m. The intent of this round is to allow the company to move aggressively into commercial markets with its RTIME technology. Earlier financing has come from Edmark, Creative Multimedia and Proxim.
709.5 Review: 2D Photographs to 3D Models by Jonathan Sunberg
A new class of software packages has been introduced lately that derives 3D models from 2D photographs. Utilizing photogrammetry, these programs can determine up to micron accuracy the camera's position and viewing direction based on 2D photographs. Once the locations are determined, the computer creates a 3D image by mathematically triangulating object points which can be seen multiple photographs.
To create a 3D model the user needs to take two or more photographs of the object, with a film, digital or video camera. WAVE used Casio's newest digital camera, the QV11(see below for more info.).
The number of photographs needed depends on the complexity of the object being converted. For simple objects, like a box or a building, three angles would suffice. But for more complex objects, like a car, several photographs may need to be taken in order to visualize all of the angles and corners.
WAVE reviewed two packages that use the technology, Photomodeler from Eos Systems, Inc. and Wireframe Express from Synthonics. Both use photogrammetry, but differ in how they calculate camera positioning. Photomodeler uses a graphical display of a 3D cube, which the user manipulates to give approximate camera locations. We found this extremely difficult to work with when dealing with larger, more complex objects, like an automobile. Wireframe Express requires the user to set calibration targets in the photo. The tutorial used a carpenter's square to get precise target points, which was easier to work with.
In both programs, once the camera locations are set and images have been loaded, the user can place points on the photograph, much like vertices on a wireframe mesh, which later will be the basis for the computer's triangular calculation. The more points used for each model will create greater accuracy and more detail, while fewer points will create a simpler model. Once these points have been placed, the computer will then create an actual wireframe mesh. The mesh can then be exported as a DXF file to most 3D or CAD software packages (3D Studio, Softimage, Alias) and textured or manipulated as a 3D object.
The first product we reviewed was PhotoModeler 2.1 from Eos Systems. Version 3.0, which is currently available in beta, will be available commercially in June or July in 1997. The new version actually promises to offer new export formats, including 3D Studio, Wavefront OBJ, VRML 2, Direct X, and 2D DXF.
PhotoModeler comes with a tutorial manual and demonstration project for the user to become accustomed to all of its capabilities. When creating objects from our own image, the task became much more difficult. When working with a complex image, determining the camera location, which seemed easy in the tutorial, actually turned out to be a daunting task. A box-like shape is maneuvered until it encompasses the object, but it doesn't take into consideration object shape differentiation.
Once the project is completed a wireframe model is created. To render the image with textures, it is best to export the model to a rendering program and work with the wireframe model from there. We found the texturing capabilities very limited.
Photomoder 3.0 will address the problem we found with 2.1 dealing with camera control. The box you were supposed to place your object was difficult to manipulate and time consuming. 3.0 has taken away manual orientation and replaced it with an automatic feature which should place the camera in the correct position with no hassle to the user. Another new feature that 3.0 will be offering allows certain functions with just one photograph. By applying four or more control points the computer can do inverse camera, mark surfaces, do ortho-production, mark surface drawings, extract and export 3D models with photo- textures, and perform distance measurements between control and surface drawing points. The new version will also allow 3D models to be exported in VRML format.
Product: PhotoModeler 2.1 Company: Eos Systems Inc. Price: $795 Platform: Win 3.1, 95, NT Minimum System: 486-33mhz, 8 MB RAM Display Requirement: 256 SVGA Features- Camera Formats: digital, video, film with CD and/or scanner Export Capabilities: DXF, VRML (3.0)
Ratings- (scale of * to *****) Documentation: *** Time to Learn: *** Ease of Use: ** Accuracy: **** Features: *** Overall Rating: 2.6
WAVE also reviewed Wireframe Express from Synthonics. We have given this program a slightly better rating than PhotoModeler mainly because it allows for automatic texturing and offers the same features (if not more) for less money.
The calibration feature on Wireframe Express was extremely easy to use compared to Photomodeler's calibrating technique. Also Wireframe Express comes with a mirror feature that allows you to get a whole image out of just half the picture. Another feature that Photomodeler does not offer is Quick-Look Rendering. This allows an original image to be mapped directly onto the wireframe the user has created and be viewed from all angles.
Wireframe Express is lacking, though, when dealing with complex objects. The program only allows two pictures to be utilized at a time in each project. Therefore if the user is dealing with a complex object he/she would have to create multiple projects and merge them together in a DXF wireframe model using Wireframe Express or another rendering program.
Product: Wireframe Express Company: Synthonics Technology Price: $395 Platform: WinNT and 95 Minimum System: 486-33mhz, 8 MB RAM Display Requirement: 256-color SVGA
Features- Camera Formats: digital, video, film with CD and/or scanner Export Capabilities: DXF
Ratings- (from * to *****) Documentation: *** Time to Learn: ** Ease of Use: ** Accuracy: *** Features: **** Overall Rating: 3
If these software packages offer more than is needed at too high a price, there are alternate solutions to photogrammetry software. FreeD from Cognicon is an Adobe Photoshop plug-in that allows 3-D model creation from just one photograph. FreeD can create multiple texture- mapped 3-D models, but unfortunately is limited to simples shapes like boxes and spheres. The program which is available for Macs, Win 3.1, 95, and NT can be exported to a VRML browser and is available for $149.
All of the images used were captured by a Casio QV11 digital camera. QV11 is Casio's newest entry in the digital camera market. The camera has a 1.8" TFT Active Matrix LCD and includes Adobe PhotoDeluxe CD. It became available in April of 1997 and has an estimated street price of $299. QV11 holds up to 96 images in flash memory with no loss of images when batteries go dead and has a resolution of 480x240 dpi.
Copyright 1997 4th WAVE Inc.
May be redistributed in full for individual readership and posted to newsgroups, Web, and FTP sites. May not be reprinted or redistributed for profit. Short quotes are permitted but must be attributed to the WAVE Report on Digital Media.