On the one hand, search engines have transformed our world. On the other hand, we are witnessing an explosion in the amount of available multimedia data. Can the text-based techniques of the former be the answer to the latter challenge?
In this talk, we first briefly present how text-based methods are recently being used to find images, audio and videos. The key is to use an associated textual description of the media.
But, what happens when this association is incorrect or does not describe the media or is not available? The second part of the seminar exhibits the automatic annotations options available and their limitations, focusing in particular on visual data.
Finally, we take an unorthodox point of view by suggesting that text may be a holy grail? To support this thesis we present some challenges and solutions.
Marcin Detyniecki is a CNRS research scientist affiliated to the
University Pierre and Marie Curie in France. He is currently visiting
researcher at Carnegie Mellon University in the United States. His area of
interest is artificial intelligence and its application multimedia challenges.
His theoretical results have been the basis of number of invited talks. His
technical solutions have highly ranked in international competitions and also
have been deployed.
He has over 50 publications in international journal and conferences. He has
also edited 3 books and a special issue of a journal. He is regular reviewer for
journals in artificial intelligence and in multimedia. He has served as program
committee of several international conferences and co-organizes AMR, the
Adaptive Multimedia Retrieval workshop, which he co-initiated.
He has been invited researcher at the University of Florence, Italy and research
staff of the University of California at Berkeley, in the
United States.