import os
import csv
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split
import cv2
import matplotlib.pyplot as plt
from keras.models import Sequential, Model
from keras.layers.core import Dense, Flatten, Activation, Dropout, Lambda
from keras.layers.convolutional import Convolution2D, Cropping2D
import numpy as np
import pandas as pd
from scipy.misc import imread
import matplotlib.pyplot as plt
from math import *
import cv2
from keras.callbacks import ModelCheckpoint
import keras
from sklearn.model_selection import train_test_split
%matplotlib inline
print('All Packages Loaded')All Packages Loaded
x = pd.read_csv('/home/carnd/test/data/driving_log.csv')
print("All Data Loaded")All Data Loaded
x.steering.plot(title='Steering data distribution', fontsize=17, figsize=(10,4), color= 'y')
plt.xlabel('frames')
plt.ylabel('Steering angle')
plt.show()
x.throttle.plot(title='Throttle data distribution', fontsize=17, figsize=(10,4), color= 'b')
plt.xlabel('frames')
plt.ylabel('Throttle')
plt.show()
x.brake.plot(title='Brake data distribution', fontsize=17, figsize=(10,4), color= 'k')
plt.xlabel('frames')
plt.ylabel('Brake')
plt.show()
x.speed.plot(title='Speed data distribution', fontsize=17, figsize=(10,4), color= 'm')
plt.xlabel('frames')
plt.ylabel('Speed')
plt.show()
samples = []
with open('/home/carnd/test/data/driving_log.csv') as csvfile:
reader = csv.reader(csvfile)
next(reader, None)
for line in reader:
samples.append(line)train_samples, validation_samples = train_test_split(samples,test_size=0.15) #simply splitting the dataset to train and validation set usking sklearn. .15 indicates 15% of the dataset is validation setimport cv2
import numpy as np
import sklearn
import matplotlib.pyplot as plt
def generator(samples, batch_size=32):
num_samples = len(samples)
while 1:
shuffle(samples)
for offset in range(0, num_samples, batch_size):
batch_samples = samples[offset:offset+batch_size]
images = []
angles = []
for batch_sample in batch_samples:
for i in range(0,3):
name = '/home/carnd/test/data/IMG/'+batch_sample[i].split('/')[-1]
main_image = cv2.cvtColor(cv2.imread(name), cv2.COLOR_BGR2RGB)
required_angle = float(batch_sample[3])
images.append(main_image)
if(i==0):
angles.append(required_angle)
elif(i==1):
angles.append(required_angle+0.2)
elif(i==2):
angles.append(required_angle-0.2)
images.append(cv2.flip(main_image,1))
if(i==0):
angles.append(required_angle*-1)
elif(i==1):
angles.append((required_angle+0.2)*-1)
elif(i==2):
angles.append((required_angle-0.2)*-1)
"""
hsv = cv2.cvtColor(main_image, cv2.COLOR_RGB2HSV)
ratio = 1.0 + 0.4 * (np.random.rand() - 0.5)
hsv[:,:,2] = hsv[:,:,2] * ratio
img = cv2.cvtColor(hsv, cv2.COLOR_HSV2RGB)
images.append(img)
if(i==0):
angles.append(required_angle*-1)
elif(i==1):
angles.append((required_angle+0.2)*-1)
elif(i==2):
angles.append((required_angle-0.2)*-1)
"""
X_train = np.array(images)
y_train = np.array(angles)
yield sklearn.utils.shuffle(X_train, y_train)
train_generator = generator(train_samples, batch_size=32)
validation_generator = generator(validation_samples, batch_size=32)Problem Statement: Find the Steering angle, given images from three cameras, i.e Left, Centre and Right.
Checklist for creating the pipeline of the NVIDIA model : https://tinyurl.com/y88xelpb
- Preprocess incoming data, centered around zero with small standard deviation
- Trim image the to train the network on the basis of the road and the steering angle only.
- Layer 1 : Convolution Layer, Number of Filters : 24, Filter size= 5x5, stride= 2x2.
- Layer 2 : Convolution Layer, Number of Filters : 36, Filter Size = 5x5, stride = 2x2.
- Layer 3 : Convolution Layer, Number of Filters : 48, Filter Size = 5x5, stride = 2x2.
- Layer 4 : Convolution Layer, Number of Filters : 64, Filter Size = 3x3, stride = 1x1.
- Layer 5 : Convolution Layer, Number of Filters : 64, Filter Size = 3x3, stride = 1x1.
- Flatten the Image from a 2D format to a row based array.
- Layer 6 : Fully Connected Layer
- A dropout layer to avoid overfitting.
- Layer 7 : Fully Connected Layer
- Layer 8 : Fully Connected Layer
model = Sequential()
model.add(Lambda(lambda x: x/127.5 - 1., input_shape=(160,320, 3), name='Normalization'))
model.add(Cropping2D(cropping=((60, 25),(0,0))))
model.add(Convolution2D(24, 5, 5, subsample=(2,2), activation='elu', name='Conv1'))
model.add(Convolution2D(36, 5, 5, subsample=(2,2), activation='elu', name='Conv2'))
model.add(Convolution2D(48, 5, 5, subsample=(2,2), activation='elu', name='Conv3'))
model.add(Convolution2D(64, 3, 3, activation='relu', name='Conv4'))
model.add(Convolution2D(64, 3, 3, activation='relu', name='Conv5'))
model.add(Flatten())
model.add(Dense(50, activation='elu', name='FC1'))
model.add(Dense(10, activation='elu', name='FC2'))
model.add(Dense(1, name='output'))
model.summary()____________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
====================================================================================================
Normalization (Lambda) (None, 160, 320, 3) 0 lambda_input_5[0][0]
____________________________________________________________________________________________________
cropping2d_5 (Cropping2D) (None, 75, 320, 3) 0 Normalization[0][0]
____________________________________________________________________________________________________
Conv1 (Convolution2D) (None, 36, 158, 24) 1824 cropping2d_5[0][0]
____________________________________________________________________________________________________
Conv2 (Convolution2D) (None, 16, 77, 36) 21636 Conv1[0][0]
____________________________________________________________________________________________________
Conv3 (Convolution2D) (None, 6, 37, 48) 43248 Conv2[0][0]
____________________________________________________________________________________________________
Conv4 (Convolution2D) (None, 4, 35, 64) 27712 Conv3[0][0]
____________________________________________________________________________________________________
Conv5 (Convolution2D) (None, 2, 33, 64) 36928 Conv4[0][0]
____________________________________________________________________________________________________
flatten_5 (Flatten) (None, 4224) 0 Conv5[0][0]
____________________________________________________________________________________________________
FC1 (Dense) (None, 50) 211250 flatten_5[0][0]
____________________________________________________________________________________________________
FC2 (Dense) (None, 10) 510 FC1[0][0]
____________________________________________________________________________________________________
output (Dense) (None, 1) 11 FC2[0][0]
====================================================================================================
Total params: 343,119
Trainable params: 343,119
Non-trainable params: 0
____________________________________________________________________________________________________
class LossHistory(keras.callbacks.Callback):
def on_train_begin(self, logs={}):
print('BEGIN TRAINING')
self.losses = []
def on_batch_end(self, batch, logs={}):
self.losses.append(logs.get('loss'))checkpoint = ModelCheckpoint('model-{epoch:03d}.h5',
monitor='val_loss',
verbose=0,
save_best_only=True,
mode='auto')batch_history = LossHistory()Early_Stopping=keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=0, verbose=0, mode='auto')Tensor_Board = keras.callbacks.TensorBoard(log_dir='./logs', histogram_freq=0, write_graph=True, write_images=False)
model.compile(loss='mse',optimizer='adam')
plot_model = model.fit_generator(train_generator, samples_per_epoch= len(train_samples), validation_data=validation_generator,callbacks = [batch_history, Tensor_Board ],nb_val_samples=len(validation_samples), nb_epoch=5,verbose=1)
model.save('model.h5')
print('Hurray! The Model has been Saved!')WARNING:tensorflow:From /home/carnd/anaconda3/envs/carnd-term1/lib/python3.5/site-packages/keras/callbacks.py:618 in set_model.: merge_all_summaries (from tensorflow.python.ops.logging_ops) is deprecated and will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.merge_all.
BEGIN TRAINING
Epoch 1/5
6720/6830 [============================>.] - ETA: 0s - loss: 0.0424
/home/carnd/anaconda3/envs/carnd-term1/lib/python3.5/site-packages/keras/engine/training.py:1569: UserWarning: Epoch comprised more than `samples_per_epoch` samples, which might affect learning results. Set `samples_per_epoch` correctly to avoid this warning.
warnings.warn('Epoch comprised more than '
6912/6830 [==============================] - 15s - loss: 0.0418 - val_loss: 0.0235
Epoch 2/5
6912/6830 [==============================] - 14s - loss: 0.0204 - val_loss: 0.0194
Epoch 3/5
6912/6830 [==============================] - 15s - loss: 0.0183 - val_loss: 0.0194
Epoch 4/5
6912/6830 [==============================] - 14s - loss: 0.0186 - val_loss: 0.0218
Epoch 5/5
6912/6830 [==============================] - 14s - loss: 0.0197 - val_loss: 0.0151
Hurray! The Model has been Saved!
def running_mean(x, N):
cumsum = np.cumsum(np.insert(x, 0, 0))
return (cumsum[N:] - cumsum[:-N]) / N
def plot_history(history):
plt.plot(running_mean(history.losses, 50))
plt.ylim(0, 0.06)
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('batches')
plt.show()
plot_history(batch_history)
plt.plot(plot_model.history['loss'])
plt.plot(plot_model.history['val_loss'])
plt.title('model mean squared error loss')
plt.ylabel('mean squared error loss')
plt.xlabel('epoch')
plt.legend(['training set', 'validation set'], loc='upper right')
plt.show()