Convert features to DATASETS format so output is same as after featuresSM.
Loads each features_[activity].mat, for each clip in each set, and merges the results
into DATASETS format. Does not try to add cuboids to the datasets as potentially this
could take too much memory.
INPUTS
nsets - number of sets
cliptypes - types of clips (cell of strings)
OUTPUTS
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
.subs - length N cell vector of sets of locations of cuboids
.desc - length N cell vector of cuboid descriptors
See also FEATURESLG, FEATURESSM