-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
162 lines (118 loc) · 4.37 KB
/
model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
# coding: utf-8
# In[105]:
import numpy as np
import tensorflow as tf
import csv
import matplotlib.pyplot as plt
import cv2
import matplotlib.image as mpimg
get_ipython().magic('matplotlib inline')
images = []
steering_angles = []
# In[199]:
images_orig = []
steering_angles_orig = []
steering_angles_all = []
for d in ["driving_data_1", "driving_data_2", "driving_data_3","driving_data_4","driving_data_5"] :
csvfile = open("/home/dk/Downloads/"+d+"/driving_log.csv")
reader = csv.reader(csvfile)
for line in reader:
angle = float(line[3])
steering_angles_all.append(angle)
#filter out 2/3 of very small steering angles
if angle > -0.001 and angle < 0.001:
if np.random.randint(0,3) is not 0:
continue
im = mpimg.imread(line[0].strip())
images_orig.append(im)
steering_angles_orig.append(angle)
#don't use left and right camera images for small steering angle images
if angle > -0.001 and angle < 0.001:
continue
im = mpimg.imread(line[1].strip())
images_orig.append(im)
steering_angles_orig.append(angle - 0.25)
im = mpimg.imread(line[2].strip())
images_orig.append(im)
steering_angles_orig.append(angle + 0.25)
print("Number of images", len(images_orig))
print("Image dimension", images_orig[0].shape)
#print("Steering angle", steering_angles_orig[0])
# In[216]:
#vizualization of sharp turns
min_steering = np.min(steering_angles_orig)
max_steering = np.max(steering_angles_orig)
idx_min = np.where(steering_angles_orig == min_steering)[0][0]
idx_max = np.where(steering_angles_orig == max_steering)[0][0]
_, (ax1,ax2) = plt.subplots(1,2, figsize=(20,10))
#ax1.subplot(1, 2, 1)
ax1.text(80,20,"steering_angle="+str(max_steering),color='white', fontsize='25')
ax1.imshow(images_orig[idx_max])
#ax2.subplot(1, 2, 2)
ax2.text(80,20,"steering_angle="+str(min_steering),color='white', fontsize='25')
ax2.imshow(images_orig[idx_min])
plt.show()
# In[205]:
#images = np.array(np.array(images_orig)[:,70:-30,])
images = np.array(images_orig)
steering_angles = np.array(steering_angles_orig)
# In[215]:
#print("Images before flipping", images.shape)
coin_toss = np.random.randint(0,3,images.shape[0])
images_flipped = np.empty([np.count_nonzero(coin_toss),160,320,3], dtype=images.dtype)
steering_angles_flipped = np.empty(np.count_nonzero(coin_toss), steering_angles.dtype)
#Flip images randomly with a probablity of 0.5
j = 0
for i, h_t in enumerate(coin_toss):
if h_t == 0:
continue
#print(i,j)
images_flipped[j] = images[i]
#print(images[i][0][0], images_flipped[j][0][0])
steering_angles_flipped[j] = steering_angles[i]
j = j+1
_, (ax1,ax2) = plt.subplots(1,2, figsize=(20,10))
ax1.imshow(images_flipped[0])
ax1.set_title('Original')
images_flipped = np.flip(images_flipped, 2)
steering_angles_flipped = -1.0*steering_angles_flipped
ax2.set_title('Flipped')
ax2.imshow(images_flipped[0])
plt.show()
#data augmentation
images_aug = np.concatenate((images,images_flipped), axis=0)
steering_angles_aug = np.concatenate((steering_angles, steering_angles_flipped), axis=0)
plt.subplot(1, 3, 1)
plt.hist(steering_angles_all,bins=20, range=(-1.0,1.0))
plt.xlabel('steering angles')
plt.subplot(1, 3, 2)
plt.hist(steering_angles_orig,bins=20, range=(-1.0,1.0))
plt.xlabel('w/ low values filtered')
plt.subplot(1, 3, 3)
plt.xlabel('w/ images flipped')
plt.hist(steering_angles_aug,bins=20, range=(-1.0,1.0))
plt.show()
print("Number of images after augmentation", len(images_aug))
# In[219]:
x = images_aug
y_ = steering_angles_aug
print(x.shape)
print(y_.shape)
from keras.models import Sequential
from keras.layers import Flatten, Dense, Lambda
from keras.layers import Convolution2D, MaxPooling2D, Cropping2D
model = Sequential()
model.add(Cropping2D(cropping=((70,30), (0,0)), input_shape=(160,320,3)))
model.add(Lambda(lambda x: (x / 255.0) - 0.5, input_shape=(60,320,3)))
model.add(Convolution2D(6,5,5,activation="relu"))
model.add(MaxPooling2D())
model.add(Convolution2D(6,5,5,activation="relu"))
model.add(MaxPooling2D())
model.add(Flatten())
model.add(Dense(120))
model.add(Dense(84))
model.add(Dense(1))
model.compile(loss='mse', optimizer='adam')
model.fit(x, y_,batch_size=32,validation_split=0.2, shuffle=True, nb_epoch=2)
model.save('model_dk.h5')
# In[ ]: