Refactor test to work with new classifier API

capstone_traffic_light_classifier/test.py

@@ -63,19 +63,11 @@ def read_and_resize_image(image_path):
     # Parse command line arguments
     args = parse_arguments()
 
-    # Parameters
-    n_classes = 4                # Namely `void`, `red`, `yellow`, `green`
-
     # Load data on which prediction will be performed
     x_batch, y_batch = load_test_data(args)
 
-    # Placeholders
-    x = tf.placeholder(dtype=tf.float32, shape=[None, args.resize_h, args.resize_w, 3])  # input placeholder
-    p = tf.placeholder(dtype=tf.float32)                                     # dropout keep probability
-    targets = tf.placeholder(dtype=tf.int32, shape=[None])
-
     # Define model
-    classifier = TrafficLightClassifier(x, targets, p, n_classes, learning_rate=1e-4)
+    classifier = TrafficLightClassifier(input_shape=[args.resize_h, args.resize_w], learning_rate=1e-4)
 
     # Add a saver to save the model after each epoch
     saver = tf.train.Saver()
@@ -86,7 +78,8 @@ def read_and_resize_image(image_path):
         saver.restore(sess, args.checkpoint_path)
 
         # Predict on loaded batch
-        prediction = sess.run(fetches=classifier.inference, feed_dict={x: x_batch, p: 1.})
+        prediction = sess.run(fetches=classifier.inference,
+                              feed_dict={classifier.x: x_batch, classifier.keep_prob: 1.})
         prediction = np.argmax(prediction, axis=1)  # from onehot vectors to labels
 
         # Qualitatively show results