A EUMSSI team from IDIAP and LIUM participated in the MediaEval 2015 Multimodal Person Discovery in Broadcast TV benchmark workshop that took place in Wurzen, Germany, on September 14 , 2015.

The participants were asked to return the names of people who can be both seen as well as heard (“Who speaks when?” and “Who appears when?”) in every shot of a collection of videos. This task represented an extension of the (now completed) French REPERE challenge, which focused on multimodal person recognition in TV broadcast.

The main challenges of this task were that the list of people was not known a priori and their names had to be discovered in an unsupervised way from media content using text overlay or speech transcripts.

The task was evaluated using information retrieval metrics, based on a posteriori collaborative annotation of the test corpus.

To ensure high quality indexes, those algorithms should also help human annotators to double-check these indexes by providing an evidence of the claimed identity (especially for people who are not yet famous).

The EUMSSI team from IDIAP and LIUM was ranked 1st among the 9 participating teams.

The teams included the best performing groups from the previous REPERE challenge (team PERCOLATTE) from Aix Marseille Université, OrangeLabs, Panasonic Silicon Valley Lab; IRISA & Inria Rennes , CNRS, Rennes; LIMSI, CNRS, France; LIG, Univ. Grenoble Alpes; Universitat Politecnica de Catalunya, Spain ; AtlantTIC Research Center; Tokyo Institute of Technology, Tokyo, Japan.

The workshop paper and poster can be found in the Publications section.