%O Report 
%F LSCOMEvent:2006
%A Kennedy, Lyndon
%T {Revision of LSCOM Event/Activity Annotations, {DTO} Challenge Workshop on Large Scale Concept Ontology for Multimedia}
%I Columbia University
%X 		The DTO-sponsored LSCOM workshop has developed a large concept lexicon for multimedia which includes approximately 			1000 concepts related events, objects, locations, people, and programs. Human subjects annotated 449 of those concepts over a corpus of 80 hours of TRECVID 2005 videos, composed of 61,901 subshots, by visually inspecting a keyframe, which is a single still image, from each subshot and making a judgment as to whether a given concept is present or absent within that keyframe. This keyframe-only approach is fast and has proven to be sufficient for many of the static concepts, such as "airplane," "car," or "person;" however, event or activity concepts, like "airplane taking off," "car driving," or "person walking," have a temporal component and are difficult for humans to judge, given only a single keyframe. It is necessary, therefore, to acquire labels based on a viewing of the video in motion to truly make a reliable annotation for the event. From the total 449 concepts that were annotated using the keyframe-based method, we selected 24 to be re-annotated using a video-based approach. These video-based annotations for 24 concepts are freely available for download, along with the keyframe-based labels for all 449 concepts. 	
%U http://www.ee.columbia.edu/dvmm/publications/06/adventtr2006-event.pdf
%8 December
%D 2006