added eval ipynb

environment.yaml

@@ -0,0 +1,4 @@
+name: carnd-capstone
+dependencies:
+	- jupyter
+	- matplotlib==2.1.2

ros/src/tl_detector/helpers.py

@@ -0,0 +1,13 @@
+import tensorflow as tf
+
+def load_graph(graph_file):
+    """Loads a frozen inference graph"""
+    graph = tf.Graph()
+    with graph.as_default():
+        od_graph_def = tf.GraphDef()
+        with tf.gfile.GFile(graph_file, 'rb') as fid:
+            serialized_graph = fid.read()
+            od_graph_def.ParseFromString(serialized_graph)
+            tf.import_graph_def(od_graph_def, name='')
+
+    return graph

ros/src/tl_detector/light_classification/tl_classifier.py

@@ -4,6 +4,20 @@
 from abc import ABCMeta, abstractmethod
 
 
+import tensorflow as tf
+from helpers import load_graph
+
+import rospkg
+import os.path as path
+
+# Model path
+r = rospkg.RosPack()
+base_path = r.get_path('tl_detector')
+SSD_INCEPTION_SIM = path.join(base_path,'light_classification/models/ssd_inception_v2_coco_retrained_sim/frozen_inference_graph.pb')
+
+# Labels dictionary
+labels_dict = {1: 'Green', 2: 'Red', 3: 'Yellow',4: 'Unknown'}
+
 class TLClassifier(object):
     """
     Base class for traffic light classifiers. The subclasses should provide implementations for the following methods:

ros/src/tl_detector/tl_detector.py

@@ -11,7 +11,6 @@
 import tf
 import yaml
 
-
 class TLDetector(object):
     def __init__(self):
 
@@ -97,10 +96,18 @@ def image_cb(self, msg):
         :param msg: image from car-mounted camera
         :type msg: Image
         """
+        if self.last_state == TrafficLight.RED:
+            # High threshold for accelerating
+            self.state_count_threshold = STATE_COUNT_THRESHOLD_DRIVE
+        else:	
+            # Low threshold for stopping
+            self.state_count_threshold = STATE_COUNT_THRESHOLD_STOP
+
         self.has_image = True
         self.camera_image_msg = msg
         light_wp, state = self.process_traffic_lights()
 
+
         # Publish upcoming red lights at camera frequency.
         # Each predicted state has to occur `self.light_classifier.get_state_count_threshold(self.last_state)` number
         # of times till we start using it. Otherwise the previous stable state is used.