featuresSM

 Feature detection and description applied to DATASETS behavior data. 

 1) First, detect features for every clip in every dataset. This leads to a large
 reduction in the size of the dataset (the original clips are discarded). 
    DATASETS = featuresSMdetect( DATASETS, par_stfeatures );

 2) Next create a descriptor, and call featuresSMpca for dimensionality reduction: 
    cubdesc = imagedesc_generate( 1, ... ); %with proper parameters
    cubdesc = featuresSMpca( DATASETS, cubdesc, kpca );

 3) Apply the descriptor to cuboids, again leading to a reduction of size of dataset:
    DATASETS = featuresSMdesc( DATASETS, cubdesc );

 See RECOGNITION_DEMO / FEATURESLG for general steps of detection / description and
 differences between this function and FEATURESLG.

 INPUTS
   DATASETS    - array of structs, should have the fields:
           .IS         - the N behavior clips
           .IDX        - length N vector of clip types
   par_stfeatures  - parameters for feature detection [see featuresSMdetect]
   cubdesc         - cuboid descriptor [see featuresSMdesc]
   kpca            - number of dimensions to reduce data to [see featuresSMpca]

 OUTUPTS
   DATASETS    - array of structs, will have additional fields:
           .IDX        - length N vector of clip types
           .ncilps     - N: number of clips
           .cubcount   - length N vector of cuboids counts for each clip clip
           .cuboids    - length N cell vector of sets of cuboids
           .subs       - length N cell vector of sets of locations of cuboids
           .desc        - length N cell vector of cuboid descriptors
   cubdesc         - output of featuresSMpca
   cuboids         - output of featuresSMpca
   
 See also FEATURESSMDETECT, FEATURESSMPCA, FEATURESSMDESC, FEATURESLG