#!/usr/bin/python3.5


from http.server import BaseHTTPRequestHandler, HTTPServer
from urllib.parse import parse_qsl

class MyRequestHandler(BaseHTTPRequestHandler):
  def do_GET(self):
    # e.g. "/?image_path=/root/mobike.jpg"                                                                                                                   
    path = self.path
    # e.g. "/root/mobike.jpg"                                                                                                                                
    image_path = parse_qsl(path[2:])[0][1] + '\n'

    # send response status code                                                                                                                              
    self.send_response(200)

    # send headers                                                                                                                                           
    self.send_header('Content-type','text/html')
    self.end_headers()

    # send message back to client, write content as utf-8 data                                                                                               
    self.wfile.write(bytes(image_path, "utf8"))
    return

def start_web_server():
  print('Starting web server...')

  server_address = ('127.0.0.1', 8080)
  httpd = HTTPServer(server_address, MyRequestHandler)
  httpd.serve_forever()


start_web_server()

./run-tensorflow-service.sh

./client.sh /root/raspberry-pi/ai/tensorflow-related/resource/test-images/mobike.jpg

import time

def run_inference_on_image(image):
  """Runs inference on an image.

  Args:
    image: Image file name.

  Returns:
    Nothing
  """
  if not tf.gfile.Exists(image):
    tf.logging.fatal('File does not exist %s', image)
  image_data = tf.gfile.FastGFile(image, 'rb').read()

  # the image used to warm-up TensorFlow model
  warm_up_image_data = tf.gfile.FastGFile('/root/tensorflow-related/test-images/ubike.jpg', 'rb').read()

  # Creates graph from saved GraphDef.
  create_graph()

  with tf.Session() as sess:
    # Some useful tensors:
    # 'softmax:0': A tensor containing the normalized prediction across
    #   1000 labels.
    # 'pool_3:0': A tensor containing the next-to-last layer containing 2048
    #   float description of the image.
    # 'DecodeJpeg/contents:0': A tensor containing a string providing JPEG
    #   encoding of the image.
    # Runs the softmax tensor by feeding the image_data as input to the graph.
    softmax_tensor = sess.graph.get_tensor_by_name('softmax:0')

    print("Warm-up start")
    for i in range(10):
      print("Warm-up for time {}".format(i))
      predictions = sess.run(softmax_tensor,
                           {'DecodeJpeg/contents:0': warm_up_image_data})

    print("Warm-up finished")

    # record the start time of the actual prediction
    start_time = time.time()

    predictions = sess.run(softmax_tensor,
                           {'DecodeJpeg/contents:0': image_data})
    predictions = np.squeeze(predictions)

    # Creates node ID --> English string lookup.
    node_lookup = NodeLookup()

    top_k = predictions.argsort()[-FLAGS.num_top_predictions:][::-1]
    for node_id in top_k:
      human_string = node_lookup.id_to_string(node_id)
      score = predictions[node_id]
      print('%s (score = %.5f)' % (human_string, score))

    print("Prediction used time:{} S".format(time.time() - start_time))

# the image used to warm-up TensorFlow model
warm_up_image_data = tf.gfile.FastGFile('/root/tensorflow-related/test-images/ubike.jpg', 'rb').read()

print("Warm-up start")
for i in range(10):
  print("Warm-up for time {}".format(i))
  predictions = sess.run(softmax_tensor,
   {'DecodeJpeg/contents:0': warm_up_image_data})

print("Warm-up finished")

# record the start time of the actual prediction
start_time = time.time()
# (中间省略)
print("Prediction used time:{} S".format(time.time() - start_time))