Kaiming he initialization for convolution layers

Question

The normal initialization for convolutional layer looks like this :

weight_scale * np.random.randn(no_of_filters,channels,kernel_size,kernel_size)

<