FOR IMMEDIATE RELEASE
Singapore, 16 November 2010 - Ever wished how you could zoom in on certain parts of a video again while searching through numerous video clips? Or seeing that favourite part of the clip that you were furiously hunting high and low for through numerous clips? Now replicate that search across 8000 videos and you can see how difficult that must be! Not so, if you are using the Known-item-Search technology from I²R.
A team from A*STAR’s Institute for Infocomm Research (I2R) managed to obtain the top scores beating 49 participating global teams and emerging best among the final 14 teams. The final 15 teams included teams from the National University of Singapore (NUS), Carnegie Mellon University, City University of Hong Kong and Dublin City University.
The main goal of TREC Video Retrieval Evaluation (TRECVID) is to promote progress in content-based analysis of and retrieval from digital video via open, metrics-based evaluation. TRECVID is a laboratory-style evaluation that attempts to model real world situations or significant component tasks involved in such situations.
The I²R team, formed in less than four months, had to search through 8000 videos and handled 300 text queries that were used for evaluation. For each query, there is only one correct answer/video. The search also covered two categories, automatic and interactive. The queries are complex, often long and may contain errors (as in the real world scenario).
Professor Lye Kin Mun, Executive Director, I²R said, “I2R has always been looking out for opportunities to benchmark our technologies through participation in global competitive events. It is through taking up such challenges that we can discover our strengths and learn from our weaknesses, eventually delivering technologies that are globally marketable and internationally acclaimed.”
In the automatic search category, I²R returned the search with mean inverted rank (See ANNEX A) of 0.454 in only 0.001 minutes per query, while in the interactive search category, we achieved mean inverted rank of 0.727, and took only 1.442 minutes per query.
The I2R team comprised of researchers from its Signal Processing (SP) and Computer Vision and Image understanding (CVIU) Departments, led by Dr Lekha Chaisorn, comprises of 7 team members: Mr Wan Kong Wah, Dr Zheng Yan-Tao, Dr Zhu Yongwei, Mr Kok Tian Shiang, Ms Tan Hui Li, Mr Fu Zixiang, and Ms Susanna Bolling.
Background
About Institute for Infocomm Research
The Institute for Infocomm Research (I²R - pronounced as i-squared-r) is a member of the Agency for Science, Technology and Research (A*STAR) family. Established in 2002, our mission is to be the globally preferred source of innovations in `Interactive Secured Information, Content and Services Anytime Anywhere’ through research by passionate people dedicated to Singapore’s economic success.
I²R performs R&D in information, communications and media (ICM) technologies to develop holistic solutions across the ICM value chain. Our research capabilities are in information technology and science, wireless and optical communications, and interactive digital media.
We seek to be the infocomm and media value creator that keeps Singapore ahead. Website: www.i2r.a-star.edu.sg
About the Agency for Science, Technology and Research (A*STAR)
The Agency for Science, Technology and Research (A*STAR) is the lead agency for fostering world-class scientific research and talent for a vibrant knowledge-based and innovation-driven Singapore. A*STAR oversees 14 biomedical sciences, and physical sciences and engineering research institutes, and nine consortia & centre, which are located in Biopolis and Fusionopolis, as well as their immediate vicinity.
A*STAR supports Singapore's key economic clusters by providing intellectual, human and industrial capital to its partners in industry. It also supports extramural research in the universities, hospitals, research centres, and with other local and international partners.
For more information about A*STAR, please visit www.a-star.edu.sg
About NIST
The National Institute of Standards and Technology (NIST) in USA develops technologies, measurement methods and standards that help U.S. companies compete in the global marketplace. The US Congress created NIST in 1901 at the start of the industrial revolution to provide the measurement and standards needed to resolve and prevent disputes over trade and to encourage standardization.
The TREC conference series is sponsored by NIST with additional support from other U.S. government agencies. The goal of the conference series is to encourage research in information retrieval by providing a large test collection, uniform scoring procedures, and a forum for organizations interested in comparing their results. In 2001 and 2002 the TREC series sponsored a video "track" devoted to research in automatic segmentation, indexing, and content-based retrieval of digital video. Beginning in 2003, this track became an independent evaluation (TRECVID) with a workshop taking place just before TREC.
For more information, please visit www.nist.gov
For media enquiries, please contact:
Mr Andrew Yap
Manager, Corporate Communications
Institute for Infocomm Research (I2R)
DID: (65) 6419 1143 Fax: (65) 6466 7716
Email: [email protected]
Annex A
Submission and Evaluation Criteria:
Given a text-only description of the video desired (i.e. a topic) and a test collection of video with associated metadata: (a) automatically return a list of up to 100 video IDs ranked by probability to be the one sough - there is no time limit on automatic searches but the elapsed time for each search - from the time the topic is presented to the system until the search result for the topic is frozen as complete - must be submitted with the system output; OR (b) interactively return the ID of the sought video and elapsed time to find it - no more than 5 minutes may elapse from the time the topic is presented to the system/searcher until the search result for the topic is frozen as complete. Ground truth will be known when the topic is created. Scoring will be automatic. Automatic runs will be scored against the ground truth using mean inverted rank at which the known item is found or equivalent. Interactive runs will be scored in terms of items found or not, elapsed time, user satisfaction.
Benchmarking Format:
Known-item search (KIS) was identified as one of the six evaluation tasks in 2010 TRECVID. The task models the situation in which someone knows of a video, has seen it before, believes it is contained in a collection, but doesn't know where to look. To begin the search process, the searcher formulates a text-only description, which captures what the searcher remembers about the target video. About 8000 videos used for the evaluation, using videos with durations between 10 seconds and 3.5 minutes, are characterized by a high degree of diversity in creator, content, style, production qualities, original collection device/encoding, language, etc - as is common in much "web video". (Source: http://www-nlpir.nist.gov/projects/tv2010/tv2010.html)
