CVPR 2001 Short Course

 

Computer Vision and the Art of Special Effects

 

Instructors: Steve Sullivan (Industrial Light and Magic), Luc Robert (REALVIZ), Irfan Essa (Georgia Institute of Technology), Steve Seitz (University of Washington), and Eugene Vendrovsky (Rhythm & Hues)

 

DESCRIPTION

 

Over the past 10 years, computer graphics has come to dominate special effects production. Despite the sophisticated tools available, however, artists still rely on essentially manual processes to complete many of their shots. As manual approaches prove too expensive, time-consuming, or inexact to meet directors' escalating demands, studios are reaching deeper into current technical research to achieve their artistic goals.

 

In this course, we discuss how computer vision techniques are impacting feature film production. After an overview of the production process, representatives from ILM and Rhythm and Hues will describe their use of vision algorithms on recent shows such as "Star Wars - The Phantom Menace", "Babe", "The Mummy", "Nutty Professor", "Pearl Harbor", "A.I.", and "Harry Potter".  Luc Robert of REALVIZ will then discuss the challenges of commercializing vision algorithms for both large and small-scale production, and Irfan Essa will describe how special effects projects are used to motivate and supplement a computer vision course at Georgia Tech. We'll close the session with discussion of problems with current algorithms as well as production issues ripe for computer vision solutions.

 

 

 

COURSE OUTLINE

 

Intro: computer vision and film production

Steve Sullivan, ILM

1:00-1:20 

 

Description of how an effects studio works, and how computer vision algorithms make their way into artists' hands.

 

 - How are studios and shows organized?

 - How is a shot put together?

 - What is the production pipeline?

 - R&D in a production environment

 - How do algorithms make their way to the screen?

 - How does vision fit in?

 - When do we buy, and when do we roll our own?

 

 

Computer vision at ILM

Steve Sullivan, ILM

1:20-2:00 

 

An overview of past and current computer vision techniques in use at Industrial Light + Magic. Topics will be discussed by deconstructing shots from the films "Star Wars - Episode I", "Mighty Joe Young", "The Mummy", "Magnolia", "Pearl Harbor", and "A.I. - Artificial Intelligence".

 

Topics covered:

 - Camera calibration

 - Matchmoving

 - Image-based modeling

 - Image-based rendering

 - Motion capture

 - On-set visualization

 - Rotoscoping

 - Object removal

 - The gap between theory and practice

 

 

 

Commercializing vision algorithms for production

Luc Robert, REALVIZ

2:00-2:45

 

Luc will speak about his experiences founding REALVIZ to commercialize computer vision algorithms for a production environment. The presentation will be illustrated with a number of production examples and technical demonstrations. Topics include:

  - Vision algorithms and film production tools: how big is the gap? 

  - Camera calibration, Matchmoving, Image-based Modeling, Retiming, Image Stitching

 

 

Break

2:45-3:00

 

Special effects and computer vision education 

Steve Seitz, U. of Washington.

Irfan Essa, Georgia Tech.

 

 

3:00-3:30

 - Steve will give an overview of "Vision for Graphics", a course he co-taught with Rick Szeliski at the University of Washington.  The course surveys many of the computer vision techniques that have applications in special effects production and computer graphics research.  The topics covered include image warping, matte extraction, motion estimation, mosaics, camera calibration, match move, shape recovery, texture analysis, and reflectance modeling.  After giving a brief overview of the course and its scope, he will present highlights from student projects.

 

 - Irfan will discuss his course at Georgia Tech which combines the technical aspects of computer vision and computer graphics, especially applied to digital video special effects production.  He'll briefly describe the topic covered, concepts discussed and show some of his students' completed projects. A similar course is also now offered at MIT. Further information and video projects are available from http://www.cc.gatech.edu/dvfx/.

 

 

 

Computer vision at Rhythm and Hues

Eugene Vendrovsky

3:30-4:15

 

Eugene will present an overview of computer vision techniques in use at Rhythm and Hues Studios. He'll discuss the use of proprietary and commercial vision algorithms at R&H, as well as present the perspective of a matchmove artist using these tools. Shot breakdowns will include material from the films "Sum of All Fears", "Harry Potter", and "Dr. Doolittle 2".

 

Topics covered:

 - 2D tracking

 - Camera matchmove

 - Object pose estimation and matchmove

 - Geometry estimation

 - Interactive tools for performing the above

 

 

Current and future problems

Steve Sullivan, ILM

4:15-5:00

 

A look at how current algorithms can fall short under real production conditions, as well as a discussion of the algorithms we'd like to see in the future.

 

Topics include:

    - Image-based modeling

    - Lighting recovery

    - Material/reflectance estimation

    - Recovery of arbitrary deforming geometry

    - "Daylight" motion capture

    - Automatic shape tracking/matting

    - Object removal/plate restoration

 

 

 

BRIEF BIOGRAPHIES:

 

STEVE SULLIVAN:

Steve is a Principal Engineer in ILM's Research and Development group, specializing in computer vision applications for film and video production. He received a PhD in Electrical Engineering in 1996 from the University of Illinois at Urbana-Champaign, with an emphasis on automatic object modeling, recognition, and surface representations. After graduation, he joined Rhythm and Hues Studios in Los Angeles to develop 3D animation and tracking software, and then moved to ILM in 1998 to focus exclusively on vision algorithms.  His work has been used on over 25 feature films including Star Wars Episode I, The Mummy, The Perfect Storm, Pearl Harbor, A.I., and Planet of the Apes.

 

 

LUC ROBERT:

Luc is Chief Technology Officer at REALVIZ (www.realviz.com), and was instrumental in the founding of the company in 1998. From an initial team of 6, REALVIZ has now grown to a total of 95 staff members, with headquarters in Sophia Antipolis, France, U.S. headquarters in San Francisco, California, and offices in Paris, England and Singapore. REALVIZ develops and commercializes image-based content creation software for professionals and amateurs alike in film, broadcasting, gaming, architecture, Internet and CAD, derived from years of research at the famous INRIA Lab in France.  REALVIZ' award-winning products enable 2D, as well as 3D artists, to cost-effectively create rich media content from pictures and video. They are currently in use by some of the world's leading production and effects houses, web design studios and architectural companies: Walt Disney Imagineering, Industrial Light and Magic, Warner Brothers Productions, Cinesite, BBC, CocaCola, Digital Domain, Buf Compagnie, Procter & Gamble, RENAULT Design, the Jim Henson Company, Cabinet Benoit amongst many others.

 

Luc received a PhD in Computer Science from the Ecole Polytechnique (Palaiseau, France) in 1993, for his work on multi-camera geometry, stereo reconstruction and camera calibration. After a one year post-doctorate at the Robotics Institute of Carnegie-Mellon University working on stereo vision for autonomous robot navigation, he joined INRIA as a Research Scientist and developed algorithms and systems for image-based modeling, image-based rendering and camera tracking. Since March 1998 he has lead the technological development of the REALVIZ' product portfolio.

 

 

STEVE SEITZ:

Steve is an assistant professor in the Department of Computer Science and Engineering at the University of Washington.  He is interested in problems in computer vision and computer graphics and the interface between these two fields.  He received the B.A. degree in computer science and mathematics at the University of California, Berkeley in 1991, and the Ph.D. degree in computer sciences at the University of Wisconsin, Madison in 1997.  Following his doctoral work, he spent one year visiting the Vision Technology Group at Microsoft Research, and subsequently two years as an assistant professor in the Robotics Institute at Carnegie Mellon University.  He joined the faculty at the University of Washington in 2001.  His current research focuses on capturing and visualizing shape, motion, and illumination of objects and environments from photographs.  He received an NSF Career Award in 2000, and was twice awarded the David Marr Prize, in 1999 and 2001, at the International Conference of Computer Vision.

 

 

IRFAN ESSA:

Irfan is an Assistant Professor at the College of Computing, Georgia Institute of Technology. There he is a member of the GVU Center, the Future Computing Environments Group, and the Aware Home Research Initiative, and has founded the Computational Perception Laboratory.  His research is in the areas of Analysis, Interpretation, and Synthesis of Video. He publishes in computer vision, graphics, AI, and HCI. He obtained his PhD/MS from MIT (1994) and also worked as a Research Scientist at MIT Media Laboratory (1994-1996) before joining Georgia Tech. http://www.cc.gatech.edu/~irfan

 

 

EUGENE VENDROVSKY:

Eugene is a Senior Software Engineer in the Rhythm & Hues Software Department, concentrating on computer vision applications for film and commercial production, image processing algorithms, and interactive computer graphics.

 

He received a PhD in Computer Graphics in 1986 from the Academy of Sciences of the USSR, with an emphasis on image processing and pattern recognition.  Eugene worked in the Academy of Sciences on academic as well as aerospace R&D on statistical detection and pattern recognition algorithms. After developing a 2d tracking production system for colorization of black and white movies, he joined R&H in 1993 as an image processing expert to develop software for various production applications. Since 1998, he has worked primarily on 3D tracking software development. His work has been used on all of Rhythm & Hues' long form projects, including Babe (which won the Academy Award for Best Visual Effects in 1995), Babe II, Ace Ventura, Nutty Professor, Flintstones, Cats & Dogs, and Dr. Doolittle II, among others.

 

 

 

CONTACT INFORMATION:

Steve Sullivan

Industrial Light + Magic

P.O. Box 2459

San Rafael, CA 94912

Tel: 415.448.3707

sullivan@ilm.com

 

 

Back to Main