)V. (2) MSA is constructed based on Attention by split the channels ofQ,K and V into h groups with each group apart ofqueries, keys,and valuesQi,Ki RN

F s,iBs,i, n = 1,2,...,S, (10) where F s is the support features extracted by a pretrained ViT. Inspired by the multiple-object tracking within a single framework [21], in which different objects are represented by various identifications (i.e., learnable vectors) for simultaneously tracking, we add extra learnable tokens tothemeanfeatures formorediscriminativeprompts.