Initial commit of Behavioural Cloning project

Personal repository creation and initial commit

model.py

@@ -0,0 +1,119 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Fri Apr  3 22:18:44 2020
+
+@author: shahid
+"""
+
+import csv, numpy as np, os, scipy.ndimage as ndimg, sklearn
+from sklearn.model_selection import train_test_split
+from sklearn.utils import shuffle
+from keras.models import Sequential, load_model
+from keras.layers import AveragePooling2D, Cropping2D, ELU, Lambda
+from keras.layers.core import Dense, Dropout, Flatten
+from keras.layers.convolutional import Conv2D
+from keras import optimizers
+
+def instantiate_model(learningRate):
+    ch, row, col = 3, 160, 320  # Original image format
+
+    model = Sequential()
+    # Pre-process incoming data, centered around zero with small standard deviation 
+    model.add(Lambda(lambda x: x/127.5 - 1.0, input_shape=(row,col,ch)))
+    # Crop the image to focus on the region of interest
+    model.add(Cropping2D(cropping=((69,25), (0,0))))
+    model.add(Conv2D(24, kernel_size=(5, 5)))
+    model.add(ELU())
+#     model.add(Conv2D(36, kernel_size=(5, 5)))
+#     model.add(ELU())
+    model.add(Conv2D(48, kernel_size=(5, 5)))
+    model.add(ELU())
+#     model.add(Conv2D(64, kernel_size=(3, 3)))
+#     model.add(ELU())
+    model.add(Conv2D(64, kernel_size=(3, 3)))
+    model.add(ELU())
+
+    model.add(Flatten())
+    model.add(Dense(100))
+    model.add(ELU())
+    model.add(Dropout(0.25))
+#     model.add(Dense(50))
+#     model.add(ELU())
+#     model.add(Dropout(0.5))
+    model.add(Dense(10))
+    model.add(ELU())
+    model.add(Dense(1))
+
+    adamOpt = optimizers.Adam(lr=learningRate)
+    model.compile(loss='mse', optimizer=adamOpt)
+
+    return model
+
+def append_data(images, angles, batchSample, steerOffset=0.2):
+    # Steering angle offsets for Centre, Left & Right images, respectively
+    offset = steerOffset*np.array([0, 1, -1])
+    for i in range(len(offset)):
+        name = '/opt/carnd_p3/data/IMG/' + batchSample[i].split('/')[-1]
+        image = ndimg.imread(name)
+        angle = float(batchSample[3]) + offset[i]
+        images.append(image)
+        angles.append(angle)
+        # Now flip the image left-to-right
+        flippedImg = np.fliplr(image)
+        images.append(flippedImg)
+        angles.append(-angle)
+
+def generator(samples, batch_size=32):
+    num_samples = len(samples)
+    while 1: # Loop forever so the generator never terminates
+        shuffle(samples)
+        for offset in range(0, num_samples, batch_size):
+            batch_samples = samples[offset:offset+batch_size]
+
+            images = []
+            angles = []
+            # Tuneable angle correction
+            steer_angle_offset = 0.2
+            for batch_sample in batch_samples:
+                append_data(images, angles, batch_sample, steer_angle_offset)
+
+            X_train = np.array(images)
+            y_train = np.array(angles)
+            yield shuffle(X_train, y_train)
+
+samples = []
+with open('/opt/carnd_p3/data/driving_log.csv') as csvfile:
+    reader = csv.reader(csvfile)
+    for line in reader:
+        samples.append(line)
+
+train_samples, validation_samples = train_test_split(samples, test_size=0.2)
+
+# Set the batch size
+batchSize = 32
+
+# Compile and train the model using the generator function
+train_generator = generator(train_samples, batch_size=batchSize)
+validation_generator = generator(validation_samples, batch_size=batchSize)
+
+print('\nLoading model ...\n\n')
+learningRate = 0.0001
+model = instantiate_model(learningRate)
+
+mdlWtsFname = os.path.abspath('./model_weights.h5')
+mdlFname = os.path.abspath('./model.h5')
+if os.path.isfile(mdlWtsFname):
+    model.load_weights(mdlWtsFname)
+
+print('Training ...\n')
+
+model.fit_generator(train_generator,
+            steps_per_epoch=np.ceil(len(train_samples)/batchSize),
+            validation_data=validation_generator,
+            validation_steps=np.ceil(len(validation_samples)/batchSize),
+            epochs=7, verbose=1)
+
+model.save(mdlFname)
+model.save_weights(mdlWtsFname)
+print('\n\nModel weights have been saved to the path below.\n' + mdlWtsFname)