End2EndDriving/Training_2D.py at master · AutoAILab/End2EndDriving · GitHub

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
import numpy as np
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten, Lambda, Cropping2D, MaxPooling2D, Conv2D, Activation, Dropout
from tensorflow.keras.applications.vgg16 import VGG16
from tensorflow.keras.layers import Input
from tensorflow.keras.models import Model
from tensorflow.keras.optimizers import Adam
import matplotlib.pyplot as plt
import tensorflow as tf
from sklearn.utils import shuffle

import cv2

train_data = np.load('data/tf_data1.npy', allow_pickle=True)

noise = 0.05
images = []
controls = []
i = 0
forward = 0
backward = 0
normal = 0
print(len(train_data))


for data in train_data:
    i = i+1
    source_image = data[0]
    new_image = cv2.resize(source_image, (200, 66))
    new_image = cv2.cvtColor(new_image, cv2.COLOR_BGR2HLS)
    new_image = cv2.GaussianBlur(new_image, (3, 3), 0)
    new_image = new_image/255 - 0.1
    control = data[1]

    # Data cleaning
    if control[1] == 0.0 or control[1] == 0.5:
        forward += 1
        images.append(new_image)
        controls.append(control)

    elif -0.3 > control[0] > 0.3 or control[1] < 0.1:  # The images with negative throttle and high steering angle are duplicated
        backward += 2
        copy_image = new_image - 0.05
        if control[0] < 0:
            copy_control = [control[0] - noise, control[1] - 0.1]
        else:
            copy_control = [control[0] + noise, control[1] - 0.1]

        images.append(new_image)
        controls.append(control)
        images.append(copy_image)
        controls.append(copy_control)
    else:
        if i % 3 == 0:
            normal += 1
            images.append(new_image)
            controls.append(control)

print("total images:", len(images))
print("forward filter", forward)
print("backward filter", backward)
print("normal images", normal)

X_train = np.array(images, dtype='float16')
y_train = np.array(controls)

X_train, y_train = shuffle(X_train, y_train)

print("the image is being trained on {} samples".format(len(X_train)))
epochs = 10
batch_size = 128

vgg_model = VGG16(weights='imagenet', include_top=False, input_shape=(66, 200, 3))  # pretrained vgg16 model with last layer chopped off.
for layer in vgg_model.layers:
    layer.trainable = False

data_input = Input(shape=(66, 200, 3))
vgg = vgg_model(data_input)
drop_1 = Dropout(0.2)(vgg)
flat_1 = Flatten()(drop_1)
dense = Dense(512, activation='relu')(flat_1)
drop_2 = Dropout(0.4)(dense)
dense_2 = Dense(256, activation='relu')(drop_2)
drop_3 = Dropout(0.4)(dense_2)
dense_3 = Dense(64, activation='relu')(drop_3)
drop_4 = Dropout(0.4)(dense_3)
prediction = Dense(2, activation='linear')(drop_4)

model = Model(inputs=data_input, outputs=prediction)

model.compile(optimizer='Adam', loss='mse')
print('training...')
model.fit(X_train, y_train, validation_split=0.2, shuffle=True, epochs=epochs, verbose=2, batch_size=batch_size)

print()
print('training complete')

model.save('models/tf_model1.h5')
print()
print('model saved')