VideoCLEF 2009: Video Analysis and Retrieval Benchmark Evaluation

VideoCLEF 2009
Video Analysis and Retrieval Benchmark Evaluation

VideoCLEF 2009 is a track of the CLEF benchmark campaign dedicated to developing and evaluating tasks involving access to video content in a multilingual environment. In 2009, organizers offer four video analysis and retrieval tasks, which will be carried out on Dutch television documentaries. Participants can approach these tasks using their own choice of methods and features. The provided video data will include speech recognition transcripts, shot boundaries, shot-level keyframes and archival metadata.

1. The Subject Classification task involves automatic tagging of videos with subject labels such as 'Music', 'History', 'Politics', and 'Museums'. This task is related to video genre classification--the subject theme labels the task uses are semantically more fine grained than genres, however. The Subject Classification task ran successfully during the VideoCLEF pilot in 2008. In 2009, this task will run on the TRECVid 2007 and 2008 collections from the Netherlands Institute for Sound and Vision. Participants are encouraged to use features derived from both the speech and visual channels.
2. The goal of the Affect and Appeal task is to move beyond the thematic content of the video and to analyze video with respect to characteristics that are important for viewers, but not related to the video topic. This task will use content from a collection of short form documentaries taken from the "Beeldenstorm" series. These are described as having "hilarious" and "moving" moments. This task comprises two subtasks, the narrative peak detection subtask, in which participants are asked to find the three funniest, most moving moments in each short documentary video, and the classification subtask, in which participants are asked to classify videos as either "popular" or "not-popular".
3. Semantic Keyframe Extraction: Keyframes or keyframe sets allow users to preview video content without playing the video. In this task, participants carry out keyframe selection using video and speech/audio features. Selected keyframes should represent the semantic content of the video, e.g., an episode of a documentary. This task will also use the "Beeldenstorm" dataset. This task builds upon the 2008 keyframe extraction task.
4. Finding Related Resources Across Languages: Given a short documentary in Dutch, participants are asked to identify English-language resources to support viewer comprehension for non-Dutch speaking viewers. This task is new in 2009. Participants will be given a number of selected time points for each short documentary and asked to link each time point to a relevant article from the English language Wikipedia. This task will also use the "Beeldenstorm" dataset.

VideoCLEF 2009 takes place on the following schedule:

• April 2009 Release of training set
• May 2009 Release of test set
• June 2009 Submission of runs
• July 2009 Evaluation
• August 2009 Working notes paper
• September 30 - October 2, 2009 CLEF Workshop