The objective of this work is set-based face recognition, i.e. to decide if
two sets of images of a face are of the same person or not. Conventionally, the
set-wise feature descriptor is computed as an average of the descriptors from
individual face images within the set. In this paper