Report from the Multimedia Grand Challenge 2010

March 2011


Authors: Cees G.M. Snoek and Malcolm Slaney

By Cees Snoek

The recent ACM Multimedia conference in Florence, Italy covered the second edition of the Multimedia Grand Challenge. An event that is designed to bring commercial industry needs to the attention of researchers. Seven industrial partners led the Multimedia Grand Challenge by identifying issues they think are important to their business and worth further study. The industrial partners pointed out that addressing these challenges would open up new business opportunities and create a richer experience for their users. In addition, they said they hoped that describing real-world challenges would allow researchers to focus on projects that have a better chance of success in the marketplace. The following 10 challenges were identified in the 2010 edition of the Multimedia Grand Challenge:

  • CeWe: Automatic theme-identification of photo sets. Provide a user with a selection of photo-book styles accommodating the user's preferences for the images in the set, and their preferred structure.
  • Google: Genre classification for video. Automatically classify user-generated videos and their metadata into genres.
  • HP: High-impact visual communication. Create a high-impact collage (and its description) that uses a set of photos to convey information across cultural boundaries.
  • Google: Interactive searching in personal diaries. Develop good schema, algorithms, and user interfaces that make multimedia diaries accessible.
  • Radvision: Multimedia-content adaptation. Adapt, in real-time, the same multimedia content for different receiving devices, in a way that is perceptually optimal for users.
  • Yahoo!: Novel image-understanding. Develop methods that move beyond simple image classification.
  • Nokia: Photo location and orientation. Derive the exact camera poses (location and orientation) for photos that are lacking location annotation.
  • Yahoo!: Robust video segmentation. Develop algorithms to automatically create narrative themes for a given video and present the content to end users in a search-engine experience.
  • 3DLife: Sports-activity analysis in camera networks. Identify the limits in terms of 2D and 3D data extraction from a low-cost camera network for sports, such as tennis matches.
  • Radvision: Videoconferencing experience. Develop new technologies and ideas to surpass the in-person meeting experience.

Together, the problems defined for the Multimedia Grand Challenge cover the full range of multimedia research. The partners contributed their problems, prize money, and data sets to the Multimedia Grand Challenge. Most importantly, they contributed judges, many of whom are directly responsible for multimedia products. In the end, 18 contributions were accepted that together covered all 10 industry problems.

All finalists for the Multimedia Grand Challenge presented their work to the panel of judges in a conference-wide event. Each finalist had three minutes to present the idea in a format that mixed an elevator pitch with a show like American Idol. After an energetic session led by master of ceremony David Ayman Shamma, the industry partners decided on the three winners. First prize of US$1,500 dollars went to Jana Machajdik, Allan Hanbury, and Julian Stöttinger who are affiliated with Vienna University of Technology and the Information Retrieval Facility, both in Austria. Machajdik presented their solution to the challenge on novel image-understanding methods. The winning idea is a system able to search for images with a requested emotional effect-such as images being cute, fearful, or sad. The second prize of US$1,000 dollars went to Wei Song, Dian Tjondronegoro, and Ivan Himawan who are all affiliated with Queensland University of Technology, Australia. Tjondronegoro presented their solution to the challenge on multimedia content adaptation. The third prize of US$500 dollars went to Julien Law-To, Gregory Grefenstette, Jean-Luc Gauvain, Guillaume Gravier, Lori Lamel, and Julien Despres who are affiliated with Exalead, LIMSI CNRS, and Vecsys Research, all in France. Law-To presented their solution to the robust video-segmentation challenge. While not all participants won prize money, everyone did go home with the message that solving real-world industry problems can be as challenging as research.

There will be another Multimedia Grand Challenge at ACM Multimedia 2011 in Scottsdale, Arizona. Hope to meet you there.

Previous Section Table of Contents Next Section