Search form

Focusing on the Face: Zemeckis Teaches Mocap at USC

Students in USC's new graduate class in performance capture, co-instructed by Robert Zemeckis, learn the importance of facial animation to coax out strong character performances. Karen Raugust reports.

Robert Zemeckis (left) with Softimage's Kevin G. Clark (center) and Thomas Kang, co-instructs a new graduate-level performance-capture class at USC's School of Cinematic Arts. © 2007 Kevin G. Clark for Softimage, Co.

As the animation industry has matured and tasks have become increasingly specialized, there's more need for specialization in the educational process as well. "Now you can have one course, not just on 3D animation but on performance capture, or even just on facial animation," says Marc Stevens, vp/gm of Softimage.

A case in point: the new graduate-level performance-capture class at USC's School of Cinematic Arts, co-instructed by the Polar Express director and Monster House producer Robert Zemeckis, who currently is working on the performance-capture film Beowulf. In February, he announced a partnership with Walt Disney Studios for the formation of a new performance capture company that will create 3D animated films using this technology, to be produced by Zemeckis and his partners and distributed by Disney.

The 15-week course, which started in January, came about when Zemeckis approached USC with the idea that students should be exposed to performance capture as a filmmaking technique. To get the program off to a good start, he told the school, he'd be willing to be involved in the first class. "He's been making films this way for the last five years," explains Eric Furie, digital systems specialist and adjunct faculty for motion-capture at USC's School of Cinematic Arts, who co-instructs the course with Zemeckis.

The 15-week course originated when Zemeckis approached USC with the idea to expose students to performance capture as a filmmaking technique. © 2007 Kevin G. Clark for Softimage, Co.

"We approached the first class with something of a directorial mindset," Furie continues. "That's where Bob's knowledge and experience come into play." The 12 grad students in the class are writing and directing a two-minute scene that relies on motion- and performance-capture techniques.

The key to performance capture is the face. "Performance capture is motion capture, but it's targeted toward capturing the whole performance of the actor, especially the face," says Furie. But facial animation -- like performance capture in general -- is complicated and difficult to teach in a single course. "It's a steep learning curve to cover the whole process in one semester," he explains. "We stripped it to its barest bones."

Adding to the challenge was the fact that the students came from diverse backgrounds, evenly divided between three of the school's divisions (production, animation and interactive), and therefore had a wide range of knowledge about 3D animation, from very little to a lot. "Half the students hadn't ever touched a 3D software package before this," Furie reports.

Eric Furie of USC's School of Cinematic Arts co-instructs the course with Zemeckis.

Facial Animation and Face Robot

The School installed Softimage|Face Robot for facial animation. "We found that it could do a good percentage of what we needed to get a taste of motion capture without a deep knowledge of 3D software packages," Furie says, noting that it allows students to quickly get to the point where they can start to tweak the animation. "We want to extract the technological drudgery and focus on the artist. That's one of the charms of Face Robot, that it allows you to do that."

"The strength of the human being is in the art and creativity," notes Thomas Kang, senior consultant at Softimage Special Projects, who co-created Face Robot and teaches in the animation division at the School of Cinematic Arts. "[Face Robot] takes care of the technology of facial rigging, so [the student] can focus on the art."

A company like ILM can put the infrastructure and team in place to do a character such as Davy Jones in Pirates of the Caribbean 2. But "to have 12 students all doing facial animation -- what army are you going to bring in?" Kang asks. "You may not be able to get 100% of Davy Jones [with Face Robot], but if you can get 70% to 80% of Davy Jones in a fraction of the time, it fills a need."

"We automate the process of getting the face set up," Stevens explains, noting that Face Robot can do in minutes what might take weeks or months for a team to set up using traditional means. Since its introduction in June 2006, the package has been used in films, commercials, music videos and especially games, where time frames are especially tight and teams small. The quick set-up also translates to more flexibility for directors, who can change their minds and do more iterations, something that is valuable in a learning environment as well.

Face Robot co-creator and USC animation teacher Thomas Kang notes that

Softimage began developing Face Robot due to requests from some of its clients, especially in the game industry, and in response to industry trends. "People in the industry had been doing a more sophisticated, realistic kind of animation," Kang says. "The expectations for the baseline quality were going up pretty dramatically." Citing characters such as Gollum in The Lord of the Rings trilogy, as well as King Kong and Davy Jones, as examples, he adds, "On the film side it was happening, but also on the game side. Especially in cinematics, and even in-game now, with the next-generation consoles."

"The process of getting a good performance out of the face of a character was a very technical process," Stevens adds. "The face drives the performance, but that was one of the areas that had gotten the least attention."

Meanwhile, one of the company's clients from the game sector, Blur Studio, was looking for new tools to keep up with clients' facial animation demands and came to Softimage Special Projects for help. "They do lots of high-quality animation in an insanely short amount of time," Kang reports, noting that existing tools "weren't scaling fast enough."

While the product started as a solution for a client, the company saw broader possibilities. After a long development process -- during which Blur helped test the product, even using beta versions in production -- Softimage introduced Face Robot to the industry at large last June.

For character generation, the school chose evolver which allows the class to create six mo-cap-ready character skeletons, with textures and skinning, without knowing how to build 3D characters. evolver images © Darwin Dimensions.

Character Generation

For the performance-capture class, USC needed a similarly simple solution for character generation. "There was no time to rig characters, and it's not a character-rigging class," Furie explains. The school chose Darwin Dimensions' evolver, which allowed the class to create six mocap-ready character skeletons, with textures and skinning, without knowing how to build 3D characters.

"It's an amazing piece of software," Furie says. "Like Face Robot, it's just jaw-dropping." evolver also worked well with Face Robot, which was important in terms of the smooth flow of the class. In fact, the process of taking the evolver mesh and making it ready for animation in Face Robot took just 15 minutes.

evolver builds characters from a virtual gene pool with interchangeable 3D facial features and body types that can be mixed and matched, tweaked and morphed while retaining the integrity of character rigging and other attributes, according to the company. evolver creator Michel Fleury, managing director of Darwin Dimensions and associate professor at the University of Québec at Montréal, notes that as characters are generated, the set-up is recalculated to the new morphology, so the skeleton, blendshapes, textures and skin weights are well-balanced for general-purpose movement. A full-feature character can be generated in about four minutes.

evolver builds characters from a virtual gene pool with interchangeable 3D facial features and body types that can be mixed and matched, tweaked and morphed while retaining the integrity of character rigging and other attributes.

"So if a course is oriented toward animation and you do not want to lose time making the character fit your needs," Fleury explains, "then you have a very fast way of creating your own personalized high-end character (your genetic code), without spending weeks of fine tweaking to prepare it for animation. The dirty work has been done automatically with our generator on our server. Eric wanted professional-quality, ready-to-animate characters, so it was a natural fit for him to use evolver."

The entire pipeline for the performance-capture class was designed to simplify the process and allow the students to focus on creating the best performances possible, Furie says. In addition to Face Robot and evolver, the class used the school's Vicon motion-capture system with Vicon IQ for control and clean-up. A Vicon plug-in was used to put the skeleton into Motion Builder for body animation. After facial animation was done in Face Robot, the facial and body data was put back into Motion Builder without using Maya or XSI, a fairly complicated step that the Softimage Special Projects team worked out. Avid Media Composer was used for editing.

Softimage's Marc Stevens views its partnership with USC as a way for the company to expose up-and-coming artists to its products. USC was the first institution in North America to have Face Robot installed.

Originally, the Vicon system was installed several years ago when the Robert Zemeckis Center for Digital Arts opened at USC's School of Cinematic Arts. It has been used for films, demos and other purposes since then, but this is the first course the School has offered that focuses specifically on mocap. For this semester only, the class gets to work with a new on-loan system that represents Vicon's latest technology.

Looking into the Future

Furie says the initial class has met and exceeded expectations, and that the School plans to move forward with a mocap curriculum. "It was a really great class, not just of its own accord, but in terms of how we teach the curriculum going forward," he says. Because the learning curve is so steep for a 15-week course, one change will be to split this class into two parts, with the first focusing on the technical side of mocap fundamentals and being a prerequisite for the second, which will deal more with the creative side, similar to the initial class. Enrollment will increase to 15 and the class will be open to interested students from all of the School's six divisions.

The partnership between the School and Softimage, like all industry-education alliances, brings benefits to both sides and will continue. "The technological hurdles for a school like ours are just too massive without help from the industry," Furie points out.

"Not a lot of schools teach facial animation because of these enormous technological hurdles," adds Kang.

For Softimage, the partnership allows the company to expose up-and-coming artists to its products. "We want them to have the latest and greatest, and to be familiar with our tools," says Stevens, who reports that USC was the first institution in North America to have Face Robot installed. A few programs in Asia are using it as well.

There also are benefits for Softimage from a product development perspective. "It opens up a lot of opportunities for artistic collaborations in one of the best film schools in the country," says Kang. He points out that Robert Zemeckis, one of the visionaries in performance capture, gave the company feedback on Face Robot and about some of the things he'd like to accomplish in the future if technology allowed it. "I can't overemphasize how valuable that is for us."

Karen Raugust is a Minneapolis-based freelance business writer specializing in animation, publishing, licensing and art. She is the author of The Licensing Business Handbook (EPM Communications).