Video Collaborative Annotation Forum: Establishing Ground-Truth Labels on Large Multimedia Datasets
Ching‐Yung Lin, Belle L. Tseng, John R. Smith · 2003
We developed a new version of The VideoAnnEx, a.k.a. IBM MPEG-7 Annotation Tool, for collaborative multimedia annotation task in a distributed environment. The VideoAnnEx assists authors in the task of annotating video sequences with MPEG-7 metadata. Each shot in the video sequence can be annotated with static scene descriptions, key object descriptions, event descriptions, and other lexicon sets. The annotated descriptions are associated with each video shot or regions in the keyframes, and are stored as MPEG-7 XML file. We proposed a forum to collaboratively annotate semantic labels to the NIST TRECVID 2003 development set. From April to July 2003, 111 researchers from 23 institutes worked together to associate 198K of ground-truth labels (433K after hierarchy propagation) to 62.2 hours of videos. This large set of valuable ground-truth data is publicly available to the research community, especially for multimedia indexing and retrieval, semantic understanding, and supervised machine learning fields. 1.