Fig. 5From: An multi-scale learning network with depthwise separable convolutionsThe comparison of MNIST accuracyBack to article page