Ability to set number of iterations

README.md

@@ -3,4 +3,4 @@
 This repository contains a number of models optimized by Intel to run machine
 learning workloads on Intel® Xeon® Scalable processors. For information on
 how to run key benchmarks, see the [README](/benchmarks)
-in the benchmarks folder. How-tos and tutorials can be found in the [docs](/docs) folder.
+in the benchmarks folder. How-tos and tutorials can be found in the [docs](/docs) folder.

docs/README.md

@@ -15,4 +15,4 @@
 * Intel-Optimized TensorFlow Serving:
     * [Image Recognition](/docs/image_recognition/tensorflow_serving/Tutorial.md) (ResNet50 and InceptionV3)
     * Object Detection (*coming soon*)
-
+

models/image_recognition/tensorflow/resnet50/fp32/eval_image_classifier_inference.py

@@ -72,12 +72,16 @@ def __init__(self):
                                  'the benchmark will use random/dummy data.',
                             dest="data_location", default=None)
 
+    arg_parser.add_argument('-s', "--steps",
+                            help='Specify the number of inference steps.',
+                            dest="steps", default=40)
+
     arg_parser.add_argument('-r', "--accuracy-only",
                             help='For accuracy measurement only.',
                             dest='accuracy_only', action='store_true')
     # parse the arguments
     self.args = arg_parser.parse_args()
-    # validate the arguements
+    # validate the arguments
     self.validate_args()
 
   def run(self):
@@ -92,7 +96,7 @@ def run(self):
 
       with tf.Session(config=config) as sess:
 
-        # convert the freezed graph to optimized graph
+        # convert the frozen graph to optimized graph
         graph_def = tf.GraphDef()
         with tf.gfile.FastGFile(self.args.input_graph, 'rb') as input_file:
           input_graph_content = input_file.read()
@@ -127,13 +131,12 @@ def run(self):
           num_remaining_images = IMAGENET_VALIDATION_IMAGES
 
         if (not self.args.accuracy_only):  # performance check
-          iteration = 0
+          step = 0
           warm_up_iteration = 10
-          total_run = 40
           total_time = 0
 
-          while num_remaining_images >= self.args.batch_size and iteration < total_run:
-            iteration += 1
+          while num_remaining_images >= self.args.batch_size and step < self.args.steps:
+            step += 1
 
             # Reads and preprocess data
             if (self.args.data_location):
@@ -149,11 +152,11 @@ def run(self):
             (predicts) = sess.run([output_tensor], feed_dict={input_tensor: image_np})
             time_consume = time.time() - start_time
 
-            print('Iteration %d: %.3f sec' % (iteration, time_consume))
-            if iteration > warm_up_iteration:
+            print('Iteration %d: %.3f sec' % (step, time_consume))
+            if step > warm_up_iteration:
               total_time += time_consume
 
-          time_average = total_time / (iteration - warm_up_iteration)
+          time_average = total_time / (step - warm_up_iteration)
           print('Average time: %.3f sec' % (time_average))
 
           print('Batch size = %d' % self.args.batch_size)