Update branch with dev's images

.gitignore

@@ -109,6 +109,7 @@ venv/
 ENV/
 env.bak/
 venv.bak/
+automec_env/
 
 # Spyder project settings
 .spyderproject
@@ -132,8 +133,9 @@ dmypy.json
 .vscode
 
 
-# images
-*.jpg
+# databases and models
+cnn/data/**
+cnn/models/**
 
 # csv
 *.csv

README.md

@@ -84,10 +84,9 @@ And add the previous line to the end of your file. Same procedure if you use zsh
 To Save press Cntrl+O , Enter . Cntrl+X to exit
 
 # Ackerman dependencies
-    sudo apt install ros-noetic-ros-controllers
-    sudo apt-get install ros-noetic-ackermann-msgs
+    sudo apt-get install ros-noetic-ros-controllers ros-noetic-ackermann-msgs
 
-# Launch woth ackerman
+# Launch with ackerman
     roslaunch ackermann_vehicle_gazebo ackermann_robot_with_arena_conversion.launch
 
 # Running the simulation enviroment

cnn/launch/training1.launch

@@ -5,6 +5,8 @@
   <arg name="batch_size" default="32"/>
   <arg name="folder" default="set1"/>
   <arg name="model" default="model1.h5"/>
+  <arg name="width" default="320"/>
+  <arg name="height" default="160"/>
 
   <!--
   <arg name="folder" default=""/>
@@ -17,6 +19,8 @@
       <param name="base_folder" value="$(arg folder)"/>
       <param name="modelname" value="$(arg model)"/>
       <param name="batch_size" value="$(arg batch_size)"/>
+      <param name="width" value="$(arg width)"/>
+      <param name="height" value="$(arg height)"/>
     </node>
   <!--</group>-->
 

cnn/launch/write.launch

@@ -2,18 +2,29 @@
 
   <arg name="image_raw_topic" default="/ackermann_vehicle/camera/rgb/image_raw"/>
   <arg name="twist_cmd_topic" default="/cmd_vel"/>
+  <arg name="vel_cmd_topic" default=""/>
   <arg name="folder" default="set1"/>
   <arg name="rate" default="30"/>
   <arg name="width" default="320"/>
   <arg name="height" default="160"/>
+  <!-- only for yaml file -->
+  <arg name="cam_height" default=""/>
+  <arg name="cam_angle" default=""/>
+  <arg name="env" default=""/>
+  <arg name="vel" default="0"/>
 
     <node name="cnn_write" pkg="cnn" type="write.py" output="screen" required="true">
       <param name="image_raw_topic" value="$(arg image_raw_topic)"/>
       <param name="twist_cmd_topic" value="$(arg twist_cmd_topic)"/>
+      <param name="vel_cmd_topic" value="$(arg vel_cmd_topic)"/>
       <param name="folder" value="$(arg folder)"/>
       <param name="rate" value="$(arg rate)"/>
       <param name="width" value="$(arg width)"/>
       <param name="height" value="$(arg height)"/>
+      <param name="cam_height" value="$(arg cam_height)"/>
+      <param name="cam_angle" value="$(arg cam_angle)"/>
+      <param name="env" value="$(arg env)"/>
+      <param name="vel" value="$(arg vel)"/>
     </node>
 
 </launch>

cnn/scripts/cnn1/training1.py

@@ -7,6 +7,9 @@
 import rospy
 from tensorflow.keras.models import load_model
 
+import yaml
+import sys
+
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
 
 def main():
@@ -24,6 +27,11 @@ def main():
     batch_yval = rospy.get_param('~batch_yval', 0)
     validation_steps = rospy.get_param('~validation_steps', 50)
 
+    image_width = rospy.get_param('~width', 320)
+    image_height = rospy.get_param('~height', 160)
+    # params only used in yaml file
+    #env = rospy.get_param('~env', 'gazebo')
+
     print('base_folder: ', base_folder)
     print('modelname: ', modelname)
     print('epochs: ', epochs)
@@ -44,6 +52,41 @@ def main():
     #print('\ndata load from ' + base_folder)
     rospy.loginfo('Train with data load from:\n %s', base_folder)
 
+    # yaml
+    if not os.path.isfile(path_data + '/info.yaml'):
+        have_dataset_yaml = False
+        # we may allow to continue processing with default data
+        print("no yaml info file found. exit.")
+        sys.exit()
+    else:
+        have_dataset_yaml = True
+
+    # dataset yaml defaults
+    ds_cam_angle = ''
+    ds_cam_height = ''
+    ds_developer = ''
+    ds_environment = ''
+    ds_frequency = 0
+    ds_image_size = ''
+    ds_linear_velocity = 0
+
+    if have_dataset_yaml:
+        with open(path_data + '/info.yaml') as file:
+            # The FullLoader parameter handles the conversion from YAML
+            # scalar values to Python the dictionary format
+            info_loaded = yaml.load(file, Loader=yaml.FullLoader)
+
+            ds_cam_angle = info_loaded['dataset']['cam_angle']
+            ds_cam_height = info_loaded['dataset']['cam_height']
+            ds_developer = info_loaded['dataset']['developer']
+            ds_environment = info_loaded['dataset']['environment']
+            ds_frequency = info_loaded['dataset']['frequency']
+            ds_image_size = info_loaded['dataset']['image_size']
+            ds_linear_velocity = info_loaded['dataset']['linear_velocity']
+
+
+    #sys.exit()
+
     # Step 2 - Vizualize and Balance data
     balanceData(data, display=True)
 
@@ -69,24 +112,78 @@ def main():
 
     print("\n" + "Create a new model from scratch? [Y/N]")
     if input().lower() == "y":
-        model = createModel()
+        model = createModel(image_width, image_height)
+        is_newmodel = True
     else:
         print('\n Model load from ' + modelname)
         model = load_model(path)
+        is_newmodel = False
+
+
+    # yaml
+    if is_newmodel:
+        imgsize_list = [image_width, image_height]
+        string_ints = [str(int) for int in imgsize_list]
+        imgsize_str = ",".join(string_ints)
+        # dataset yaml defaults
+        model_developer = os.getenv('automec_developer')
+        model_image_size = imgsize_str
+        model_frequency = ds_frequency
+        model_linear_velocity = ds_linear_velocity
+        model_environment = ds_environment
+        model_cam_angle = ds_cam_angle
+        model_cam_height = ds_cam_height
+        model_cnn_number = '1'
+    else:
+        with open(path + '_info.yaml') as file:
+            # The FullLoader parameter handles the conversion from YAML
+            # scalar values to Python the dictionary format
+            info_loaded = yaml.load(file, Loader=yaml.FullLoader)
+
+            model_developer = info_loaded['model']['developer']
+            model_image_size = info_loaded['model']['image_size']
+            model_frequency = info_loaded['model']['frequency']
+            model_linear_velocity = info_loaded['model']['linear_velocity']
+            model_environment = info_loaded['model']['environment']
+            model_cam_angle = info_loaded['model']['cam_angle']
+            model_cam_height = info_loaded['model']['cam_height']
+            model_cnn_number = info_loaded['model']['cnn_number']
+
 
     model.summary()
 
     # Step 9 -Training
-    history = model.fit(batchGen(xTrain, yTrain, batch_xtrain, batch_ytrain), steps_per_epoch=steps_per_epoch, epochs=epochs,
-                        validation_data=batchGen(xVal, yVal, batch_xval, batch_yval), validation_steps=validation_steps)
+    history = model.fit(batchGen(xTrain, yTrain, batch_xtrain, batch_ytrain, image_width, image_height), steps_per_epoch=steps_per_epoch, epochs=epochs,
+                        validation_data=batchGen(xVal, yVal, batch_xval, batch_yval, image_width, image_height), validation_steps=validation_steps)
 
     # Step 10 - Saving and plotting
     #model.save('models_files/' + modelname)
     model.save(path)
+
     print('\n Model Saved to ' + path)
     print('\n Model Saved to ' + modelname)
     rospy.loginfo('Model Saved to: %s', path)
 
+    # yaml
+    info_data = dict(
+
+        model = dict(
+            developer = os.getenv('automec_developer'),
+            image_size = model_image_size,
+            frequency = model_frequency,
+            linear_velocity = model_linear_velocity,
+            environment = model_environment,   
+            cam_height = model_cam_height,
+            cam_angle = model_cam_angle,
+            cnn_number = model_cnn_number
+        )
+    )
+
+    with open(path + '_info.yaml', 'w') as outfile:
+        yaml.dump(info_data, outfile, default_flow_style=False)
+
+    rospy.loginfo('yaml Saved')
+
     plt.plot(history.history['loss'])
     plt.plot(history.history['val_loss'])
     plt.legend(['Training', 'Validation'])

cnn/scripts/cnn1/utils.py

@@ -123,12 +123,12 @@ def augmentImage(imgPath, steering):
     return img, steering
 
 
-def preProcessing(img):
+def preProcessing(img, imgwidth, imgheight):
     # Cropping Region of intrest, Ajust with Gazebo and use Andre Code in the Future
     # img = img[60:135, :, :]
     img = cv2.cvtColor(img, cv2.COLOR_RGB2YUV)  # For better jornalization
     img = cv2.GaussianBlur(img, (3, 3), 0)
-    img = cv2.resize(img, (320, 160))  # NIVIDA uses 200x66
+    img = cv2.resize(img, (imgwidth, imgheight))  # NIVIDA uses 200x66
     img = img/255
 
     return img
@@ -139,7 +139,7 @@ def preProcessing(img):
 # plt.show()
 
 
-def batchGen(imagesPath, steeringList, batchSize, trainFlag):
+def batchGen(imagesPath, steeringList, batchSize, trainFlag, imgwidth, imgheight):
     # Creates a batch and applies augmentation
     while True:
         imgBatch = []
@@ -154,18 +154,18 @@ def batchGen(imagesPath, steeringList, batchSize, trainFlag):
             else:
                 img = mpimg.imread(imagesPath[index])
                 steering = steeringList[index]
-            img = preProcessing(img)
+            img = preProcessing(img, imgwidth, imgheight)
 
             imgBatch.append(img)
             steeringBatch.append(steering)
 
         yield(np.asarray(imgBatch), np.asarray(steeringBatch))
 
 
-def createModel():
+def createModel(imgwidth, imgheight):
     model = Sequential()
     model.add(Convolution2D(24, (5, 5), (2, 2),
-                            input_shape=(160, 320, 3), activation='elu'))
+                            input_shape=(imgheight, imgwidth, 3), activation='elu'))
     model.add(Convolution2D(36, (5, 5), (2, 2), activation='elu'))
     model.add(Convolution2D(48, (5, 5), (2, 2), activation='elu'))
     model.add(Convolution2D(64, (3, 3), activation='elu'))

cnn/scripts/cnn2/train2a_transfer_learning.py

@@ -101,7 +101,7 @@ def main():
 
                 # Change image and store the same angle, with some distortion, both in image and in angle (to increase robustness)
                 # flipped
-                center_image_temp = np.expand_dims(cv2.flip(center_image, 1), axis=2)
+                #center_image_temp = np.expand_dims(cv2.flip(center_image, 1), axis=2)
 
                 # images.append(center_image_temp)
                 # angles.append(-float(line[1]))
@@ -118,8 +118,21 @@ def main():
         csvfile.close()
 
 
-        left_to_straight_ratio = total_straight_angles/total_left_angles
-        right_to_straight_ratio = total_straight_angles/total_right_angles
+        if total_left_angles == 0:
+            left_to_straight_ratio = None
+        else:
+            left_to_straight_ratio = total_straight_angles/total_left_angles
+
+        if total_right_angles == 0:
+            right_to_straight_ratio = None
+        else:
+            right_to_straight_ratio = total_straight_angles/total_right_angles
+
+
+
+
+
+
 
         #print('angles are: ' + str(list(angles)))