Now that we've enhanced the classifier to load random images, we'll start with choosing these random images with the exact size and shape of our training/testing images. We'll need to add placeholders for these user-provided images, so we'll add the following lines in the appropriate locations:

 tf_random_dataset = tf.placeholder(tf.float32, shape=(1, 
  image_size, image_size, num_channels),  
 name='RANDOM_DATA')random_prediction =  
 tf.nn.softmax(nn_model(tf_random_dataset, 
  weights, biases))

Next, we will grab the image provided by the user via the following command-line parameter and run our session on the image:

We will follow almost the exact sequence as we did earlier. Running a test file through the script using the -e switch will yield an extra output, as follows:

    The prediction is: 2 

Voila! We just classified an arbitrary image.