FCN - Create the layers for a fully convolutional network. Build skip-layers using the vgg layers.

layers.py

def layers(vgg_layer3_out, vgg_layer4_out, vgg_layer7_out, num_classes):
   
    # Use a shorter variable name for simplicity
    layer3, layer4, layer7 = vgg_layer3_out, vgg_layer4_out, vgg_layer7_out

    # Apply 1x1 convolution in place of fully connected layer
    fcn8 = tf.layers.conv2d(layer7, filters=num_classes, kernel_size=1, name="fcn8")

    # Upsample fcn8 with size depth=(4096?) to match size of layer 4 so that we can add skip connection with 4th layer
    fcn9 = tf.layers.conv2d_transpose(fcn8, filters=layer4.get_shape().as_list()[-1],
    kernel_size=4, strides=(2, 2), padding='SAME', name="fcn9")

    # Add a skip connection between current final layer fcn8 and 4th layer
    fcn9_skip_connected = tf.add(fcn9, layer4, name="fcn9_plus_vgg_layer4")

    # Upsample again
    fcn10 = tf.layers.conv2d_transpose(fcn9_skip_connected, filters=layer3.get_shape().as_list()[-1],
    kernel_size=4, strides=(2, 2), padding='SAME', name="fcn10_conv2d")

    # Add skip connection
    fcn10_skip_connected = tf.add(fcn10, layer3, name="fcn10_plus_vgg_layer3")

    # Upsample again
    fcn11 = tf.layers.conv2d_transpose(fcn10_skip_connected, filters=num_classes,
    kernel_size=16, strides=(8, 8), padding='SAME', name="fcn11")

    return fcn11