Tensorflow's 'Label_image.py' in a loop increasingly slow

Question

I am trying to create an image classification program on a Raspberry Pi 3 using TensorFlow and a modified version of label_image.py.

I'm using a MobileNet model which I obtained from here. At the beginning, classifications take around 3 seconds, but this increases over time (up to over 7 seconds after 10 minutes), and I can't figure out why this is happening.

Here is the code within my loop:

while True:

  startTime = datetime.now()

  t = read_tensor_from_image_file(file_name,
                              input_height=input_height,
                              input_width=input_width,
                              input_mean=input_mean,
                              input_std=input_std)

  input_name = "import/" + input_layer
  output_name = "import/" + output_layer
  input_operation = graph.get_operation_by_name(input_name);
  output_operation = graph.get_operation_by_name(output_name);

  results = sess.run(output_operation.outputs[0],
                  {input_operation.outputs[0]: t})
  results = np.squeeze(results)

  top_k = results.argsort()[-5:][::-1]
  labels = load_labels(label_file)
  for i in top_k:
    print(labels[i], results[i])

  print(datetime.now() - startTime)

The TensorFlow session is started before the loop along with the loading of the graph.

I'm using Python 3.4.2 and TensorFlow 1.3.0.

I found another question on StackOverflow with the same issue. I tried the solution posted there but I get errors stating "AttributeError: 'Tensor' object has no attribute 'endswith'".

Answer 1

I found a solution here which works for me. By adding a with tf.Graph().as_default(): around the body of the read_tensor_from_image_file() function, my classifications now take around 1.20s even after 30 mins.

so my read_tensor_from_image_file() function looks like this:

def read_tensor_from_image_file(file_name, input_height=192, input_width=192,
                            input_mean=0, input_std=255):
  with tf.Graph().as_default():
    input_name = "file_reader"
    output_name = "normalized"
    file_reader = tf.read_file(file_name, input_name)
    if file_name.endswith(".png"):
      image_reader = tf.image.decode_png(file_reader, channels = 3,
                                   name='png_reader')
    elif file_name.endswith(".gif"):
      image_reader = tf.squeeze(tf.image.decode_gif(file_reader,
                                              name='gif_reader'))
    elif file_name.endswith(".bmp"):
      image_reader = tf.image.decode_bmp(file_reader, name='bmp_reader')

    else:
      image_reader = tf.image.decode_jpeg(file_reader, channels = 3,
                                    name='jpeg_reader')
    float_caster = tf.cast(image_reader, tf.float32)
    dims_expander = tf.expand_dims(float_caster, 0);
    resized = tf.image.resize_bilinear(dims_expander, [input_height, 
    input_width])
    normalized = tf.divide(tf.subtract(resized, [input_mean]),[input_std])

    result = sess.run(normalized)

    return result