lab-11-2-mnist_deep_cnn The output of Fully-connected layer1 about deeplearningzerotoall

[kkweon]

Hi, your question is ambiguous. So I will try my best to answer your question.

Why does it have 625 output?

Because in the previous line,

# W4.shape = (128 * 4 * 4, 625)
W4 = tf.get_variable("W4", shape=[128 * 4 * 4, 625],
                     initializer=tf.contrib.layers.xavier_initializer())

So the shape of X * W4 will be (?, 625) where X is the previous output of shape (?, 128 * 4 * 4)

Why is it set to 625?

It doesn't have to be 625 because it's a hyperparameter. It just happens to be in this case.