The Star challenge is a race to build the next-generation of multi-media search engines. A US$100,000 cash prize awaits the winners of this world-class competition. The objective of the search challenge is to encourage participation from international teams to develop new, interesting and practical search techniques.

The voice search challenge consists of three subtasks, as outlined below:

1. Search by IPA
The query is given in International Phonetic Alphabet (IPA), the task is to retrieve all segments that contain the query IPA sequence regardless of its spoken languages;
2. Search by example
The query is an utterance spoken by different speakers, the task is to retrieve all segments that contain the query word/phrase/sentence regardless of its spoken languages;
3. Search for recurrent voice segments In the voice archive, certain word/phrase/sentences are repeated more than once in the content. The task is to extract all recurrent segments which are at least 15 seconds in length. No query is given in this case. The number of unique recurrent segments for each document is given.

The video search challenge consists of three subtasks, as outlined below:

1. Search by (Single) Query Image
The query is provided in the form of a single image. The task is to retrieve all visually similar segments. Note that the similarity is at the perceptual level. That is, the expected results should contain video segments that contain images that look similar to the query image, as opposed to the video content being semantically similar. There are 30 query image types.
2. Search by Video Shot
The query is a short video shot (<10sec). The task is to retrieve video shots that are perceptually similar to the query video clip. Note that compared to VT1, there is now additional motion information in the query video shot and the matching criteria should also take into consideration similarity in the motion trajectory. There are 30 query shot types.
3. Object/Scene Categorization
A list of object/scene classes will be defined. For each class, a set of images/videos depicting the objects/scenes will be provided. The participants are expected to develop a model of the class by visually learning on the sample images/video. Then, given a new, unseen test set of images/video the task is to categorize the test set into the classes. Note that the set of test queries will necessarily be a very large set in the order of 10K queries. Also, about 10% of these queries will not belong to any of the object/ scene classes, and the desired output result is a “Reject” class. There are 30 object/scene classes.

For more details, get the White Paper here: The Star Challenge 2008

Or read this good summary from the Get Updated blog.

A global competition for evolving next generation search engine:

Singapore’s A*STAR (Agency for Science, Technology and Research) has received an overwhelming response to its global competition with an idea of conceiving technologies for a rich, dynamic media search engine. The new search engine would be smart enough for identifying text, audio & video containing any word – even if that particular word, or relevant search term, hasn’t been tagged yet in the internet material. Millions of online search engine users across the globe stand to benefit from such advanced technologies, which will help them in navigating the abundant material on the Internet.

Termed ‘The Star Challenge 2008’, this eight-month contest has a prize of USD100,000. The Star Challenge 2008 is open to people all over the world. The top software engineers, researchers and amateur search media enthusiasts taking part in it have embarked on an enchanting discovery of the new search technology. The contest is part of the Fusionopolis, Singapore’s science and technology powerhouse, which looks to shape or alter the lifestyles & economy of the future.

Slated to be held in October 2008, Fusionopolis will attract over 1500 talented people from a diverse scientific domains integrating their amazing capabilities for jointly creating the complex technologies. Developing such advanced search capabilities will radically alter the way individuals interact with large amount of multi-media information, thus creating seamless, accessible platforms across different online communities.

This entry was posted on Friday, February 29th, 2008 at 11:22 am and is filed under News. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.