How to get the output of the fully connected layer from CNN in Tensorflow?

Question

I want to extract CNN activations from the fully connected layer in a convolution neural network using tensorflow. In the following post, a user asked that question:

How to extract activation from CNN layers using tensorflow?

And the answer is this:

sess = tf.InteractiveSesssion()

full_connected = ....
value_of_fully_connected = sess.run(fully_connected,feed_dict={your_placeholders_and_values)

However, in my code, the fully connected layer is separated from the tf.session() . and I have this code for the function that computes the convolutions:

def conv_net(x, weights, biases):  

    conv1 = conv2d(x, weights['wc1'], biases['bc1'])
    conv1 = maxpool2d(conv1, k=2)

    conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
    conv2 = maxpool2d(conv2, k=2)

    conv3 = conv2d(conv2, weights['wc3'], biases['bc3'])
    conv3 = maxpool2d(conv3, k=2)


    # Fully connected layer
    fc1 = tf.reshape(conv3, [-1, weights['wd1'].get_shape().as_list()[0]])
    fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
    fc1 = tf.nn.relu(fc1)

    out = tf.add(tf.matmul(fc1, weights['out']), biases['out'])
    return out

And then the prediction:

pred = conv_net(x, weights, biases)

cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))

optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

And here is the training:

with tf.Session() as sess:
    sess.run(init) 
    train_loss = []
    test_loss = []
    train_accuracy = []
    test_accuracy = []
    summary_writer = tf.summary.FileWriter('./Output', sess.graph)
    for i in range(training_iters):
        for batch in range(len(train_X)//batch_size):
            batch_x = train_X[batch*batch_size:min((batch+1)*batch_size,len(train_X))]
            batch_y = train_y[batch*batch_size:min((batch+1)*batch_size,len(train_y))]    
            # Run optimization op (backprop).
                # Calculate batch loss and accuracy
            opt = sess.run(optimizer, feed_dict={x: batch_x,
                                                              y: batch_y})
            loss, acc = sess.run([cost, accuracy], feed_dict={x: batch_x,
                                                              y: batch_y})
        print("Iter " + str(i) + ", Loss= " + \
                      "{:.6f}".format(loss) + ", Training Accuracy= " + \
                      "{:.5f}".format(acc))
        print("Optimization Finished!")

        # Calculate accuracy for all 10000 mnist test images
        test_acc,valid_loss = sess.run([accuracy,cost], feed_dict={x: test_X,y : test_y})
        train_loss.append(loss)
        test_loss.append(valid_loss)
        train_accuracy.append(acc)
        test_accuracy.append(test_acc)
        print("Testing Accuracy:","{:.5f}".format(test_acc))
    summary_writer.close()

As you can see, the fully connected layer is inside the function conv_net() , and I cannot seem to have access to that from inside tf.session() code.

I need to have access to that fully connected layer so I can use the answer in the post above. How can I do that?

Answer 1

in python, you can return a list of output from a function. So, I would do something like:

def conv_net(x, weights, biases):  

    conv1 = conv2d(x, weights['wc1'], biases['bc1'])
    conv1 = maxpool2d(conv1, k=2)

    conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
    conv2 = maxpool2d(conv2, k=2)

    conv3 = conv2d(conv2, weights['wc3'], biases['bc3'])
    conv3 = maxpool2d(conv3, k=2)


    # Fully connected layer
    fc1 = tf.reshape(conv3, [-1, weights['wd1'].get_shape().as_list()[0]])
    fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
    fc1 = tf.nn.relu(fc1)

    out = tf.add(tf.matmul(fc1, weights['out']), biases['out'])
    return [out,fc1]

and when you want to get the outputs, you do:

pred, fcn = conv_net(x, weights, biases)

When you wanna see the result inside the session, do:

fcn_evaluated = sess.run(fcn)
print(fcn_evaluated)