Semantic Model Vectors Features for the 2010 Trecvid Multimedia Event Detection(MED) task (http://www.nist.gov/itl/iad/mig/med10.cfm).


The data is subdivided into two folders: one for development and one for evaluation.

Each folder contains 3 files:

1. framesFOLDERNAME.tdb : is the index with names of the videos and frames from which the Semantic Model Vectors were extracted.
                          
                          Each line follows the convention:  [index] [videoName/videoName_frameNumber.png]* * * *
              
                          For example:  20 HVC1040/HVC1040_7.png* * * *
                          means that the feature vector found in modelVector280FOLDERNAME.tdb at index 20 represents frame number 7 in video HVC1040
 						  
2. modelVector280FOLDERNAME.tdb : contains the actual Semantic Model Vectors Features. 
                                  Each line represents the Semantic Model Vector descriptor for a frame, has 282 entries, and follows the convention

                                  [index] [dimensions] [SM1] [SM2] ... [SM280]
								  
                                  - index is the same as in framesFOLDERNAME.tdb, to match descriptors to frames
                                  - dimensions indicates how many entries follow until the end of the line (it is always 280)
                                  - SMx represents the output of Semantic Model x on the specified frame, folowing the order specified in ModelList.txt.
                                    For example, SM12 is the prediction of the Semantic Model for the concept 'Bicycle'

3. annotationsFOLDERNAME.tdb : contains the annotation of every video clip with respect to the three event categories of Trecvid MED10: 
                               
                               1 - "Assembling a shelter"
                               2 - "Batting a run"
                               3 - "Making a cake"
                               4 - "random", meaning a video representing none of the previous three events							   
								


If you use this data in your work, please cite

Michele Merler, Bert Huang, Lexing Xie, Gang Hua and Apostol Natsev, "Semantic Model Vectors for Complex Video Event Recognition", 
to appear in IEEE Transactions on Multimedia, Special Issue on Object and Event Classification in Large-Scale Video Collections, 2012

@article{SMV_TMM,
   author = {Michele Merler and Bert Huang and Lexing Xie and Gang Hua and Apostol Natsev},
   title = {Semantic Model Vectors for Complex Video Event Recognition},
   journal = {IEEE Transactions on Multimedia},
   note  = {Special issue on Object and Event Classification in Large-Scale Video Collections },
   year = {2012}
}


For any questions contact mmerler@cs.columbia.edu