How to achieve this look in a convolutional network?

Question

Neural networks

I would like to get the following result:

I'm trying to repeat someone else's mesh, but for some reason everything turns out differently on the fully_connected layer.
The Reshape layer appears and there is no multiplication by 40 in the fully_connected layer

conv7 = conv2d(conv6, weights['wc7'], biases['bc7'])

    # Fully connected layer
    fc1 = tf.contrib.layers.fully_connected(conv7, int(biases['bc8'].get_shape()[0]), activation_fn=tf.nn.relu, weights_initializer=None)

    sfm = tf.nn.softmax(
        fc1,
        axis=None,
        name="labels_softmax",
        dim=None
    )

Answer 1

In theory, you need to insert reshape before fully connected to bring (None, 10, 4, 32) -> (None, 1280)