dcgan = DCGAN()
train_images = <images batch>
losses = dcgan.loss(train_images)
train_op = dcgan.train(losses)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    for step in range(FLAGS.max_steps):
        _, g_loss_value, d_loss_value = sess.run([train_op, losses[dcgan.g], losses[dcgan.d]])
    # save trained variables

dcgan = DCGAN()
images = dcgan.sample_images()

with tf.Session() as sess:
    # restore trained variables

    generated = sess.run(images)
    with open('<filename>', 'wb') as f:
        f.write(generated)