Gravar-mail: Learning view invariant recognition with partially occluded objects