Understanding number of parameters in Keras Conv2D layer

Question

My first layer is:

model.add(tf.keras.layers.Conv2D(filters=32, kernel_size=3, padding="same", activation="relu", input_shape=[32, 32, 3]))

And the number of parameters in the Model summary table:

    Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_4 (Conv2D)            (None, 32, 32, 32)        896  

As per my understanding, the number of parameters must be :

(No of filters) X (Number of parameters in Kernel)

i.e. in my case ==> 32 X (3 X 3) = 288

But its 896. How it comes to 896?

Thanks

Answer 1

Number of parameters in Keras Conv2D layer is calculated using the following equation:

number_parameters = out_channels * (in_channels * kernel_h * kernel_w + 1)  # 1 for bias

So, in your case,

in_channels = 3
out_channels = 32
kernel_h = kernel_w = 3
number_parameters = 32(3*3*3 + 1) = 896