Fig. 7From: An multi-scale learning network with depthwise separable convolutionsThe comparison of SVHN accuracyBack to article page