In [1]:

    

#This is the works for Udacity Self-driving-car-nd Term1 Project 3.
%matplotlib inline
import os, sys, random
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import pandas as pd


    

In [2]:

    

# Print iterations progress
def print_progress(iteration, total):
    """
    Call in a loop to create terminal progress bar
    
    Parameters
    ----------
        
    iteration : 
                Current iteration (Int)
    total     : 
                Total iterations (Int)
    """
    str_format = "{0:.0f}"
    percents = str_format.format(100 * (iteration / float(total)))
    filled_length = int(round(100 * iteration / float(total)))
    bar = '█' * filled_length + '-' * (100 - filled_length)

    sys.stdout.write('\r |%s| %s%%' % (bar, percents)),

    if iteration == total:
        sys.stdout.write('\n')
    sys.stdout.flush()


    

In [3]:

    

#Load sample data
Folder_path = "./dataset/behavioral-cloning/sample_data/"
Img_path = Folder_path + "IMG/"
df = pd.read_csv(Folder_path + "driving_log.csv")

df['left'] = df['left'].str.replace(' IMG/','./dataset/behavioral-cloning/sample_data/IMG/')
df['right'] = df['right'].str.replace(' IMG/','./dataset/behavioral-cloning/sample_data/IMG/')
df['center'] = df['center'].str.replace('IMG/','./dataset/behavioral-cloning/sample_data/IMG/')


    

In [4]:

    

Folder_path = "./dataset/behavioral-cloning/train_data_new/"
Img_path = Folder_path + "IMG/"
df = pd.read_csv(Folder_path + "driving_log.csv", names=['center', 'left', 'right', 'steering', 'gas', 'brake', 'speed'])

df['left'] = df['left'].str.replace('/home/jaeoh2/Tools/linux_sim/','./dataset/behavioral-cloning/')
df['right'] = df['right'].str.replace('/home/jaeoh2/Tools/linux_sim/','./dataset/behavioral-cloning/')
df['center'] = df['center'].str.replace('/home/jaeoh2/Tools/linux_sim/','./dataset/behavioral-cloning/')


    

In [5]:

    

df


    

    
Out[5]:







  

    

      

      
center
      
left
      
right
      
steering
      
gas
      
brake
      
speed
    
  
  

    

      
0
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19027
    
    

      
1
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19027
    
    

      
2
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19024
    
    

      
3
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19024
    
    

      
4
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19026
    
    

      
5
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19024
    
    

      
6
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.000000
      
1.000000
      
0.0
      
30.19023
    
    

      
7
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.030075
      
1.000000
      
0.0
      
30.18975
    
    

      
8
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.030075
      
1.000000
      
0.0
      
30.18998
    
    

      
9
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.030075
      
1.000000
      
0.0
      
30.19011
    
    

      
10
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.015038
      
1.000000
      
0.0
      
30.19029
    
    

      
11
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.007519
      
1.000000
      
0.0
      
30.19028
    
    

      
12
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.007519
      
1.000000
      
0.0
      
30.19029
    
    

      
13
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.007519
      
1.000000
      
0.0
      
30.19030
    
    

      
14
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.007519
      
1.000000
      
0.0
      
30.19032
    
    

      
15
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.015038
      
1.000000
      
0.0
      
30.19021
    
    

      
16
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.18995
    
    

      
17
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.18988
    
    

      
18
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.18998
    
    

      
19
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19009
    
    

      
20
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19015
    
    

      
21
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19020
    
    

      
22
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.052632
      
1.000000
      
0.0
      
30.19020
    
    

      
23
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19034
    
    

      
24
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19031
    
    

      
25
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19030
    
    

      
26
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19029
    
    

      
27
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19027
    
    

      
28
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
1.000000
      
0.0
      
30.19027
    
    

      
29
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.030075
      
1.000000
      
0.0
      
30.19032
    
    

      
...
      
...
      
...
      
...
      
...
      
...
      
...
      
...
    
    

      
13947
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.075188
      
0.304401
      
0.0
      
22.91770
    
    

      
13948
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.037594
      
0.132153
      
0.0
      
22.94271
    
    

      
13949
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
-0.015038
      
0.000000
      
0.0
      
22.79306
    
    

      
13950
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.000000
      
0.169356
      
0.0
      
22.70264
    
    

      
13951
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.000000
      
0.359982
      
0.0
      
22.79327
    
    

      
13952
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.007519
      
0.169378
      
0.0
      
22.86276
    
    

      
13953
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.015038
      
0.000000
      
0.0
      
22.73524
    
    

      
13954
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.015038
      
0.081686
      
0.0
      
22.60064
    
    

      
13955
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.015038
      
0.252669
      
0.0
      
22.59212
    
    

      
13956
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.022556
      
0.075609
      
0.0
      
22.60300
    
    

      
13957
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.047589
      
0.0
      
22.42578
    
    

      
13958
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.237558
      
0.0
      
22.39575
    
    

      
13959
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.122955
      
0.0
      
22.45437
    
    

      
13960
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.088035
      
0.0
      
22.34171
    
    

      
13961
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.323415
      
0.0
      
22.35437
    
    

      
13962
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.303414
      
0.0
      
22.50396
    
    

      
13963
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.060810
      
0.0
      
22.50279
    
    

      
13964
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.033942
      
0.0
      
22.36448
    
    

      
13965
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.219306
      
0.0
      
22.33097
    
    

      
13966
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.203121
      
0.0
      
22.42202
    
    

      
13967
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.000000
      
0.0
      
22.32722
    
    

      
13968
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.155529
      
0.0
      
22.23268
    
    

      
13969
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.257238
      
0.0
      
22.27766
    
    

      
13970
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.006210
      
0.0
      
22.22781
    
    

      
13971
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.108972
      
0.0
      
22.10771
    
    

      
13972
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.037594
      
0.223569
      
0.0
      
22.15195
    
    

      
13973
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.045113
      
0.035820
      
0.0
      
22.10633
    
    

      
13974
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.045113
      
0.142032
      
0.0
      
21.96889
    
    

      
13975
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.045113
      
0.288480
      
0.0
      
22.00568
    
    

      
13976
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
./dataset/behavioral-cloning/train_data_new/IM...
      
0.045113
      
0.051621
      
0.0
      
21.99993
    
  

13977 rows × 7 columns

In [6]:

    

#data balancing refer from https://navoshta.com/end-to-end-deep-learning/
balanced = pd.DataFrame()
bins = 200
bin_n = 200

start = 0
for end in np.linspace(0, 1, num=bins):
    df_range = df[(np.absolute(df.steering) >= start) & (np.absolute(df.steering) < end)]
    range_n = min(bin_n, df_range.shape[0])
    if range_n > 0:
        balanced = pd.concat([balanced, df_range.sample(range_n)])
    start = end
balanced.to_csv(Folder_path + 'driving_log_balanced.csv', index=False)


    

In [7]:

    

df.hist(column=['steering'], bins=bins, figsize=(10,5))


    

    
Out[7]:





array([[<matplotlib.axes._subplots.AxesSubplot object at 0x7fe3ab61a320>]], dtype=object)






    

In [8]:

    

balanced.hist(column=['steering'], bins=bins, figsize=(10,5))


    

    
Out[8]:





array([[<matplotlib.axes._subplots.AxesSubplot object at 0x7fe3aae7e940>]], dtype=object)






    

In [9]:

    

num_examples = len(df)
num_balanced = len(balanced)
print("Number of examples : {}\nNumber of balanced : {}".format(num_examples, num_balanced))


    

    




Number of examples : 13977
Number of balanced : 10775

In [10]:

    

import random
import os, sys
import csv
from skimage.transform import rotate, warp, ProjectiveTransform, AffineTransform
import cv2


    

In [11]:

    

def random_translate(X, steer, intensity=1):
    delta = 15.* intensity
    rand_delta = random.uniform(-delta, delta)
    translate_matrix = AffineTransform(translation=(rand_delta, 0))
    X = warp(X,translate_matrix)
    steer += rand_delta*0.01
    return X, steer


    

In [12]:

    

#test
ran_idx = random.randint(0, num_examples)
test_img = mpimg.imread(df.center[ran_idx])
test_str = float(df.steering[ran_idx])
test_rt_img, test_rt_str = random_translate(test_img, test_str)
plt.subplot(2,1,1), plt.imshow(test_img), plt.title(round(test_str,2)), plt.axis('off')
plt.subplot(2,1,2), plt.imshow(test_rt_img), plt.title(round(test_rt_str,2)), plt.axis('off')


    

    
Out[12]:





(<matplotlib.axes._subplots.AxesSubplot at 0x7fe34b30fa90>,
 <matplotlib.image.AxesImage at 0x7fe34b2409b0>,
 <matplotlib.text.Text at 0x7fe34b2200b8>,
 (-0.5, 319.5, 159.5, -0.5))






    

In [13]:

    

def image_flip(X, steer):
    img_fliped = np.fliplr(X)
    str_fliped = -steer
    return img_fliped, str_fliped


    

In [14]:

    

#test
test_fl_img, test_fl_str = image_flip(test_img, test_str)
plt.subplot(2,1,1), plt.imshow(test_img), plt.axis('off'), plt.title(round(test_str,2))
plt.subplot(2,1,2), plt.imshow(test_fl_img), plt.axis('off'), plt.title(round(test_fl_str,2))


    

    
Out[14]:





(<matplotlib.axes._subplots.AxesSubplot at 0x7fe34b1e7cc0>,
 <matplotlib.image.AxesImage at 0x7fe349977ef0>,
 (-0.5, 319.5, 159.5, -0.5),
 <matplotlib.text.Text at 0x7fe3499565f8>)






    

In [15]:

    

def random_rotate(X,intensity=3):
    delta = np.radians(intensity) #deg to rad
    rand_delta = random.uniform(-delta, delta) #rotate range
    rotate_matrix = AffineTransform(rotation=rand_delta)
    X = warp(X, rotate_matrix)
    return X


    

In [16]:

    

def random_brightness(image):
    #Convert 2 HSV colorspace from RGB colorspace
    hsv = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
    #Generate new random brightness
    rand = random.uniform(0.3,1.0)
    hsv[:,:,2] = rand*hsv[:,:,2]
    #Convert back to RGB colorspace
    new_img = cv2.cvtColor(hsv, cv2.COLOR_HSV2RGB)
    return new_img


    

In [17]:

    

def random_shadow(image):
    #refer from : https://chatbotslife.com/using-augmentation-to-mimic-human-driving-496b569760a9
    top_y = 320*np.random.uniform()
    top_x = 0
    bot_x = 160
    bot_y = 320*np.random.uniform()
    hls = cv2.cvtColor(image, cv2.COLOR_RGB2HLS)
    shadow_mask = 0*hls[:,:,1]
    X_m = np.mgrid[0:image.shape[0],0:image.shape[1]][0]
    Y_m = np.mgrid[0:image.shape[0],0:image.shape[1]][1]
    shadow_mask[((X_m-top_x)*(bot_y-top_y) -(bot_x - top_x)*(Y_m-top_y) >=0)]=1
    if np.random.randint(2) == 1:
        random_bright = 0.5
        cond1 = shadow_mask==1
        cond0 = shadow_mask==0
        if np.random.randint(2)==1:
            hls[:,:,1][cond1] = hls[:,:,1][cond1]*random_bright
        else:
            hls[:,:,1][cond0] = hls[:,:,1][cond0]*random_bright
        
    img = cv2.cvtColor(hls, cv2.COLOR_HLS2RGB)
    
    return img


    

In [18]:

    

import math
import cv2
height=64
width=64
def preprocess_image(X):
    X = cv2.resize(X[60:140,:], (width,height), interpolation=cv2.INTER_AREA)
    #X = cv2.cvtColor(X, cv2.COLOR_RGB2YUV)
    
    return X


    

In [19]:

    

#test
plt.figure(dpi=200)
for i in range(5):
    test_resize_img = random_brightness(test_img)
    test_resize_img = random_shadow(test_resize_img)
    test_resize_img = preprocess_image(test_resize_img)
    plt.subplot(1,5,i+1), plt.imshow(test_resize_img), plt.axis('off')


    

In [21]:

    

center = df.center.tolist()
left = df.left.tolist()
right = df.right.tolist()
steering = df.steering.tolist()
leftsteering = (df.steering + 0.25).tolist()
rightsteering = (df.steering - 0.25).tolist()


    

In [20]:

    

center = balanced.center.tolist()
left = balanced.left.tolist()
right = balanced.right.tolist()
steering = balanced.steering.tolist()
leftsteering = (balanced.steering + 0.25).tolist()
rightsteering = (balanced.steering - 0.25).tolist()


    

In [22]:

    

from sklearn.model_selection import train_test_split
from sklearn.utils import shuffle
X_train = center + left + right
y_train = steering + leftsteering + rightsteering

X_train, X_valid, y_train, y_valid = train_test_split(X_train, y_train, test_size=0.2)


    

In [23]:

    

def transform_images(X, y):
    X = random_brightness(X)
    #X = random_shadow(X)
    #X = random_rotate(X)
    if np.random.randint(2) == 0:
        X, y = image_flip(X, y)
    #X,y = random_translate(X, y, intensity=1)
    X = preprocess_image(X)
    return X,y


    

In [24]:

    

#test
plt.figure(dpi=150)
for i in range(5):
    test_img = mpimg.imread(X_train[0])
    plt.subplot(5,5,i+1),plt.imshow(transform_images(test_img, y_train[0])[0]),plt.axis('off')
    plt.subplot(5,5,i+6),plt.imshow(transform_images(test_img, y_train[0])[0]),plt.axis('off')
    plt.subplot(5,5,i+11),plt.imshow(transform_images(test_img, y_train[0])[0]),plt.axis('off')
    plt.subplot(5,5,i+16),plt.imshow(transform_images(test_img, y_train[0])[0]),plt.axis('off')
    plt.subplot(5,5,i+21),plt.imshow(transform_images(test_img, y_train[0])[0]),plt.axis('off')
    plt.subplot(5,5,1),plt.imshow(test_img),plt.axis('off')


    

In [25]:

    

def get_transform_images(Xs, ys, n_each=10):
        
    X_arr = []
    y_arr = []
    
    
    #Parallel(n_jobs=nprocs)(delayed(preprocess_image)(imfile) for imfile in Xs)
    
    for i, (x, y) in enumerate(zip(Xs,ys)):
        for _ in range(n_each):
            img = mpimg.imread(x)
            img_trf, label_trf = transform_images(img, y)
            X_arr.append(img_trf)
            y_arr.append(label_trf)
        
        print_progress(i+1, len(Xs))
        
    X_arr = np.asarray(X_arr, dtype=np.float32)
    y_arr = np.asarray(y_arr, dtype=np.float32)
    
    return X_arr, y_arr


    

In [26]:

    

X_trf, y_trf = get_transform_images(X_train, y_train, n_each=10)


    

    




 |████████████████████████████████████████████████████████████████████████████████████████████████████| 100%

In [27]:

    

X_trf_val, y_trf_val = get_transform_images(X_valid, y_valid, n_each=1)


    

    




 |████████████████████████████████████████████████████████████████████████████████████████████████████| 100%

In [28]:

    

X_trf.shape, y_trf.shape, X_trf_val.shape, y_trf_val.shape


    

    
Out[28]:





((167720, 64, 64, 3), (167720,), (8387, 64, 64, 3), (8387,))

In [29]:

    

print("Saving preprocessed Train images")
#np.save(Folder_path + "TrainX.npy", X_trf)
#np.save(Folder_path + "Trainy.npy", y_trf)
#np.save(Folder_path + "ValidX.npy", X_trf_val)
#np.save(Folder_path + "Validy.npy", y_trf_val)
print("Saving Done")


    

    




Saving preprocessed Train images
Saving Done

In [30]:

    

X_train, y_train, X_valid, y_valid = X_trf, y_trf, X_trf_val, y_trf_val


    

In [39]:

    

from keras.models import Model, Sequential
from keras.layers import Input, Conv2D, MaxPool2D
from keras.layers import Dense, Dropout, Flatten, Cropping2D
from keras.optimizers import adam


    

In [40]:

    

from keras.callbacks import ModelCheckpoint, EarlyStopping
from keras import backend as K
from keras.layers import Lambda


    

In [41]:

    

#Model Configurations
img_height = 64
img_width = 64
img_ch = 3

f_size = 3
learning_rate = 1e-2
activation = 'elu'


    

In [46]:

    

def build_model():
    model = Sequential()
    #model.add(Lambda(lambda x: x/255. -0.5, input_shape=(img_height, img_width, img_ch)))
    model.add(Conv2D(16,(3,3), activation=activation, padding='valid' , input_shape=(img_height, img_width, img_ch)))
    model.add(MaxPool2D(pool_size=(2,2)))
    model.add(Conv2D(32,(3,3), activation=activation, padding='valid'))
    model.add(MaxPool2D(pool_size=(2,2)))
    model.add(Conv2D(64,(3,3), activation=activation, padding='valid'))
    model.add(MaxPool2D(pool_size=(2,2)))
    
    model.add(Flatten())
    model.add(Dense(500, activation=activation))
    model.add(Dropout(0.5))
    model.add(Dense(100, activation=activation))
    model.add(Dropout(0.25))
    model.add(Dense(20, activation=activation))
    model.add(Dropout(0.25))
    model.add(Dense(1))
    
    model.summary()
    #model = make_parallel(model,2)
    model.compile(loss='MSE', optimizer='adam', metrics=['accuracy'])
    
    return model


    

In [47]:

    

model = build_model()
model_checkpoint = ModelCheckpoint('model.h5', monitor='loss', save_best_only=True)
model_earlystopping = EarlyStopping(monitor='loss')


    

    




_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_6 (Conv2D)            (None, 62, 62, 16)        448       
_________________________________________________________________
max_pooling2d_7 (MaxPooling2 (None, 31, 31, 16)        0         
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 29, 29, 32)        4640      
_________________________________________________________________
max_pooling2d_8 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 12, 12, 64)        18496     
_________________________________________________________________
max_pooling2d_9 (MaxPooling2 (None, 6, 6, 64)          0         
_________________________________________________________________
flatten_3 (Flatten)          (None, 2304)              0         
_________________________________________________________________
dense_9 (Dense)              (None, 500)               1152500   
_________________________________________________________________
dropout_7 (Dropout)          (None, 500)               0         
_________________________________________________________________
dense_10 (Dense)             (None, 100)               50100     
_________________________________________________________________
dropout_8 (Dropout)          (None, 100)               0         
_________________________________________________________________
dense_11 (Dense)             (None, 20)                2020      
_________________________________________________________________
dropout_9 (Dropout)          (None, 20)                0         
_________________________________________________________________
dense_12 (Dense)             (None, 1)                 21        
=================================================================
Total params: 1,228,225
Trainable params: 1,228,225
Non-trainable params: 0
_________________________________________________________________

In [48]:

    

import numpy as np
#Folder_path = "./dataset/behavioral-cloning/sample_data/"
#Folder_path = "./dataset/behavioral-cloning/train_data/"

#X_train = np.load(Folder_path + "TrainX.npy")
#y_train = np.load(Folder_path + "Trainy.npy")
#X_valid = np.load(Folder_path + "ValidX.npy")
#y_valid = np.load(Folder_path + "Validy.npy")


    

In [49]:

    

model.fit(X_train, y_train, batch_size=128, epochs=10, verbose=1, shuffle=True,
          validation_data=(X_valid, y_valid),
         callbacks=[model_checkpoint, model_earlystopping])


    

    




Train on 167720 samples, validate on 8387 samples
Epoch 1/10
167720/167720 [==============================] - 45s - loss: 1.3035 - acc: 0.0050 - val_loss: 0.0700 - val_acc: 0.0088
Epoch 2/10
167720/167720 [==============================] - 45s - loss: 0.2393 - acc: 0.0069 - val_loss: 0.0528 - val_acc: 0.0089
Epoch 3/10
167720/167720 [==============================] - 46s - loss: 0.1488 - acc: 0.0079 - val_loss: 0.0422 - val_acc: 0.0089
Epoch 4/10
167720/167720 [==============================] - 46s - loss: 0.1014 - acc: 0.0087 - val_loss: 0.0390 - val_acc: 0.0091
Epoch 5/10
167720/167720 [==============================] - 46s - loss: 0.0694 - acc: 0.0090 - val_loss: 0.0313 - val_acc: 0.0089
Epoch 6/10
167720/167720 [==============================] - 47s - loss: 0.0489 - acc: 0.0091 - val_loss: 0.0317 - val_acc: 0.0092
Epoch 7/10
167720/167720 [==============================] - 48s - loss: 0.0360 - acc: 0.0092 - val_loss: 0.0265 - val_acc: 0.0093
Epoch 8/10
167720/167720 [==============================] - 48s - loss: 0.0286 - acc: 0.0092 - val_loss: 0.0210 - val_acc: 0.0092
Epoch 9/10
167720/167720 [==============================] - 48s - loss: 0.0237 - acc: 0.0092 - val_loss: 0.0196 - val_acc: 0.0092
Epoch 10/10
167720/167720 [==============================] - 48s - loss: 0.0214 - acc: 0.0093 - val_loss: 0.0168 - val_acc: 0.0092






    
Out[49]:





<keras.callbacks.History at 0x7fe07c166400>

In [38]:

    

with open('model.json', 'w') as file:
        file.write(model.to_json())


    

In [ ]: