Supplementaryto" IntermediatePrototypeMining TransformerforFew-ShotSemanticSegmentation "

We observe that the performance of our method does not achieve a significant improvement under the5-shot setting. Wearguethatthisisreasonable because asthenumberofthe support images increases, their internal diversity probably increases. Thus, the ability of our method to correct the support-query deviation will be weakened. Wefurthershowsomequalitative results of stacking different IPMT layers in Figure 2 to prove the effectiveness of our proposed iterative prototype mining scheme.