Crystal Palace Animation Production Notes


Product

A 3-minute animation of the interior of the Crystal Palace was produced for the Museum of London, a collaborator on the “Monuments and Dust” project. The goal of the animation was to impress upon the viewer the massive scale of the building and the character of the structure and space. Delivered as a DVD, the animation accompanies a large physical model of the building in the museum’s permanent World City Gallery.

Process

The model used in the animation is an adaptation of a highly detailed kit-of-parts created from paper plans of the prototype components. One of the originally intended uses for the kit-of-parts was a super-computer-based, radiance lighting simulation. The simulation was put on hold before completion. Two years later, the parts were rebuilt in a simplified form to create animation on desktop computers. In rendering the animation, Phong shading, rather than Ray Tracing was used to keep render times under 2 minutes per frame. As a result, there are no reflections or transmissive lighting effects in the animation. Using the radiance lighting simulation test renderings as a subjective guide, light bloom was added to the final animation as a blurred composite of the transparency of the raw animation. This blur and glow effect somewhat makes up for the lack of reflections and true radiance lighting.


Fly through workflow

  1. Build 3D model
  2. Organize data for export to animation program
  3. Export data to animation program
  4. Create image maps
  5. Assign material properties to model components
  6. Place lights and set lighting parameters
  7. Render test frames and make revisions as needed
  8. Define animation path
  9. Render test animations and make revisions as needed
  10. Render final animation frames
  11. Add blur, glow, titles and transitions
  12. Create menus for DVD interface
  13. Compress animation to MPEG-2 format
  14. Test DVD functionality
  15. Compile and burn DVD
  16. Backup data


Resources

Electric Image was selected, as it is one of the few packages that will render a 3.3 million, polygon model in a reasonable time, on our hardware.

Modeling, texture mapping, lighting, animation setup, and composite work were done on an Apple Power Mac G4 with the following specifications:

DVD authoring and web movie compression were done on an Apple Power Mac G4 with the following specifications:

Rendering was done on six Apple Power Mac G4 computers with the following specifications:

A VST 25GB Firewire drive was used to shuttle data from the authoring computer to the rendering and compression computers. The project would have taken significantly longer without this device.

Statistics from the model and animation

Lessons learned

A project needs to be well-planned before work begins: Our original plan was to create a computer model that would enable various kinds of visualizations, fly-throughs, and derivative visual products, as well as enabling analysis of its design principles, and providing scholars with a better understanding of how the building worked as a piece of structural engineering, and a better sense of the aesthetic experience of being in the building. The second goal was to demonstrate (to the NSF supercomputing centers and others) that humanities computing could make use of the most advanced computing hardware and software, and could generate compute-intensive problems. When we started the project we had no idea we would render a three-minute animation using desktop computers. If that had been one of the original goals, we would have created a far less detailed model.

During the 2 year break in the modeling part of the project, rendering software went through a significant evolution. These advancements helped the project come to completion but introduced significant additional time for retraining. The break in the project added additional time to reorient to the intricacies of the model.

This kind of work is best done in teams: 3-D visualization is sufficiently complex as to overwhelm the casual user and frustrate the average user. This is true of most complex computing tasks such as programming, database development, web design, or interactive media design. The animation production pipeline has many steps, including modeling, image-map creation, lighting, animation, rendering, and post processing. In commercial environments these tasks are divided among individuals with significant expertise in their area of production. A manager, for example, oversees the process and ensures that all the parts come together as desired. It is possible for one individual to gain enough experience in each of these areas to pull together a complete project but it would be difficult to match the skill level of a small team.

Sufficient hardware resources are key for animation work: In commercial animation environments, individual frames are distributed to a farm of render machines and then returned to a host machine for automated assembly into an animation. Render farms are often used for both low-resolution test animations and final frames. Without a render farm, the visual feedback loop of the animator and the resulting revision cycle can be very slow. Iteration through a fast feedback loop and revision cycle is key to producing high quality animation. We utilized computers from the neighboring Digital Media Lab to render the final animation and a single, local, desktop machine for test animations. The DML machines were available for rendering between 10:00 pm and 10:00 am Monday through Thursday and 6:00 pm Friday to noon Saturday. The configuration of these public access machines changed almost daily, requiring more preparation than a dedicated render farm.

The animation was manually split into a number of segments that would complete during the available time and copied to individual machines via a Firewire drive. The rendered animation segments were retrieved on the same drive. Moving gigabytes of data across 10-Base-T or even 100-Base-T Ethernet can take hours. Gigabyte-sized file transfers with a Firewire drive, on the other hand, take minutes. Gigabit Ethernet and/or Firewire drives are essential technologies for dealing with large animation files.

The total data of the animation portion of this project approaches 20GB. Backing up 20GB of data to CD-R discs is not desirable, particularly if individual files exceed the 620MB CD capacity. DVD-R technology is not only suited to delivery of video but also making backups of the components used to create it. Roxio’s Toast 5 offers a straightforward solution for burning DVD-R data discs. DVD-R discs use the UDF format. The difference between a DVD-ROM disc and DVD-Video disc is only the data that is stored on the disc. This is one reason why the UDF format was adopted. Discs can be used for different purposes without requiring a change in format. The UDF format is robustly supported on the Mac but poorly supported in Windows 98 and Windows 2000.

Conclusion

3D visualization is a powerful tool and can substantially increase a project’s impact and educational value. The benefits are significant enough to justify full-time staff assigned solely to the task. In a humanities research environment, 3D visualization will stretch technical resources to the limit and take longer than in the commercial realm where a small team rather than an individual might handle the project. Due to the rapid pace of change in hardware and software and the training time needed to accommodate the changes, visualization projects should be continuous from start to finish. Changing the hardware or software used for visualization in the middle of a project can lead to significant delays in rebuilding or otherwise converting the model. Starting with a clear goal and the resources to reach that goal will reduce the time investment and improve the quality of the final product.

Chris Jessee,12.17.01
cj8n@virginia.edu