Call for Participation in the 2014 Retrieving Diverse Social Images Task survey
MediaEval 2014 Multimedia Benchmark
http://www.multimediaeval.org/mediaeval2014/
*About MediaEval*
MediaEval (http://www.multimediaeval.org) is a benchmarking initiative dedicated to evaluating new algorithms for multimedia access and retrieval. It emphasizes the 'multi' in multimedia and focuses on human and social aspects of multimedia tasks. MediaEval attracts participants who are interested in multimodal approaches to multimedia involving, e.g., speech recognition, visual analysis, music and audio analysis, user-contributed information (tags, comments, tweets), viewer affective response, social networks, geo-coordinates and non-linear video access.
The MediaEval 2014 season kicks off with the MediaEval 2014 Survey. The survey is used to collect your opinion about which tasks should be offered by the MediaEval multimedia benchmark in 2014:
https://www.surveymonkey.com/s/mediaeval2014
The survey will take you about 5 minutes if you fill in only the main questions. There are 13 main questions, which are multiple choice questions that collect your opinion on each of the 13 tasks that have been proposed for MediaEval 2014. However, we encourage you to answer the additional questions on the tasks that most interest you ó your answers contribute to decisions than are made about the design and implementation of the tasks.
The MediaEval 2014 task list will be finalized in mid February and sign up for participation will open at the beginning of March. Please be sure to fill your email address in on the first page of the survey if you would like to receive an email when sign up opens.
Our goal is to have the survey filled out by as many researchers as possible in the next three weeks ó please pass the survey link along to colleagues in the field of multimedia who might be interested.
Note that the deadline for results submissions this year will be early to mid-September and the workshop will be held in October in Barcelona.
*About 2014 Retrieving Diverse Social Images Task*
This task is a follow-up of last year's edition. The task addresses the problem of result diversification in the context of social photo retrieval. The task this year is build around the same use case scenario as in 2013 (we use a tourist use case). The participating systems are expected, given a ranked list of location photos retrieved from Flickr using text and GPS queries, to refine the results by providing a set of images that are in the same time relevant to the query and provide a diversified summary of it (initial results are typically noisy and redundant). The refinement and diversification process will be based on the social metadata associated with the images and on the visual characteristics of the images.
This year's novelty will be in exploring the effect of user annotation credibility on relevance and diversity. Credibility is determined as an automatic estimation of the quality (correctness) of a particular user's tags (a specifically designed dataset will be used to train this measure). Participants will be encouraged to exploit this provided credibility estimation in addition to classical retrieval techniques.
Moreover, another novelty for this year will consist of introducing a more adequate diversification scenario by considering the annotation of up to 300 images per location - compared to 150 last year.
Target communities involve both machine and human media analysis such as image retrieval (text, vision, multimedia communities), re-ranking, relevance feedback, crowd-sourcing and automatic geo-tagging. To solve the challenge, participants are free to choose from any approaches, from human oriented, machine-based to hybrid machine-human; as well as to take advantage of any additional data sources, e.g., the Internet. To encourage participation of groups from different research areas, additional resources such as general purpose visual descriptors and textual models will be provided for the entire collection. Evaluation of performance is going to be carried out by comparison with human generated ground truth.
*Task organizers*
Bogdan Ionescu, LAPI, University Politehnica of Bucharest, Romania;
Adrian Popescu, CEA LIST, France;
Mihai Lupu, Vienna University of Technology, Austria;
Henning M¸ller, University of Applied Sciences Western Switzerland in Sierre, Switzerland.
Thank you for your interest and input.
On behalf of the task organizers,
Mihai Lupu
Vienna University of Technology
|