A.5 Trialseeds FortheresultsofTable1andFigure2,weusedthreeseeds {0,1,2}inDomainNetlibrary,whichisthesame withtheoriginalpaper. FortheresultsofTable2weusedthreeseeds{10,11,12}aswell

For visualization, we split backbone networks into (1) Conv (blue): pure convolution (ResNet18, ResNet50, and BiT), (2) ViT (orange): purevision transformer(ViT-B16 andViT-L16), and(3)ViT+Conv(green): hybrid ofconvolution andtransformer (HViTandDeiT).