DepthwiseConv2D outputs normally when dilation_rate is 0

[Shuyan Liao]

repo code:

from keras.engine import Model, Input
layer = keras.layers.DepthwiseConv2D(kernel_size=3, padding="valid", strides=3, dilation_rate=0, data_format="channels_first")
x = Input(batch_shape=(1, 32, 32, 16))
y = layer(x)
model = Model(x, y)