Retina enhanced descriptors for spatio-temporal concept detection.