-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathModel.py
executable file
·191 lines (168 loc) · 7.7 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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
import os
import h5py
from create_data_files import ImageProcessor
import tensorflow as tf
import numpy as np
import pickle
data_dir = './Data/processed'
def load_data(dataset='cropped', normalised=True, grey=False):
""""Returns the dataset chosen by parameters, only use one of normalised/grey/plain"
Keyword Arguments:
dataset {string} -- [Name of dataset to load] (default: {cropped})
normalised {bool} -- [Normalised dataset] (default: {True})
grey {bool} -- [greyscale dataset] (default: {False})
"""
if normalised:
directory = 'normalised'
elif grey:
directory = 'grey'
image_processor = ImageProcessor(data_dir)
train_data, train_labels = image_processor.load_data('train', dataset)
test_data, test_labels = image_processor.load_data('test', dataset)
"""
extra_data, extra_labels = image_processor.load_data('extra', dataset)
train_data = train_data + extra_data
train_labels = train_labels + extra_labels
"""
return train_data, train_labels, test_data, test_labels
def create_model(X, y):
input_layer = tf.keras.Input(shape=(X.shape[1:]))
cl1 = tf.keras.layers.Conv2D(filters=48, kernel_size=(5, 5),
padding='same', input_shape=X.shape[1:],
activation='relu', use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(input_layer)
bnl1 = tf.keras.layers.BatchNormalization(axis=-1)(cl1)
mpl1 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(2, 2), padding='same')(bnl1)
dpl1 = tf.keras.layers.Dropout(rate=0.5)(mpl1)
# second layer
cl2 = tf.keras.layers.Conv2D(filters=64, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl1)
bnl2 = tf.keras.layers.BatchNormalization(axis=-1)(cl2)
mpl2 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(1, 1), padding='same')(bnl2)
dpl2 = tf.keras.layers.Dropout(rate=0.5)(mpl2)
# third layer
cl3 = tf.keras.layers.Conv2D(filters=128, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl2)
bnl3 = tf.keras.layers.BatchNormalization(axis=-1)(cl3)
mpl3 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(2, 2), padding='same')(bnl3)
dpl3 = tf.keras.layers.Dropout(rate=0.5)(mpl3)
# Fourth layer
cl4 = tf.keras.layers.Conv2D(filters=160, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl3)
bnl4 = tf.keras.layers.BatchNormalization(axis=-1)(cl4)
mpl4 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(1, 1), padding='same')(bnl4)
dpl4 = tf.keras.layers.Dropout(rate=0.5)(mpl4)
# fifth layer
cl5 = tf.keras.layers.Conv2D(filters=192, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl4)
bnl5 = tf.keras.layers.BatchNormalization(axis=-1)(cl5)
mpl5 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(2, 2), padding='same')(bnl5)
dpl5 = tf.keras.layers.Dropout(rate=0.5)(mpl5)
# sixth layer
cl6 = tf.keras.layers.Conv2D(filters=192, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl5)
bnl6 = tf.keras.layers.BatchNormalization(axis=-1)(cl6)
mpl6 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(1, 1), padding='same')(bnl6)
dpl6 = tf.keras.layers.Dropout(rate=0.5)(mpl6)
# seventh layer
cl7 = tf.keras.layers.Conv2D(filters=192, kernel_size=(5, 5),
padding='same', activation='relu',
use_bias=True,
kernel_regularizer=tf.keras.regularizers.l2(
0.001))(dpl6)
bnl7 = tf.keras.layers.BatchNormalization(axis=-1)(cl7)
mpl7 = tf.keras.layers.MaxPool2D(pool_size=(
2, 2), strides=(2, 2), padding='same')(bnl7)
dpl7 = tf.keras.layers.Dropout(rate=0.5)(mpl7)
# 1st fully connected layer
fl1 = tf.keras.layers.Flatten()(dpl7)
dl1 = tf.keras.layers.Dense(3072, activation='relu')(fl1)
dpl8 = tf.keras.layers.Dropout(rate=0.5)(dl1)
# 2nd fully connected layer
dl2 = tf.keras.layers.Dense(3072, activation='relu')(dpl8)
dpl9 = tf.keras.layers.Dropout(rate=0.5)(dl2)
# output layer
output_digit1 = tf.keras.layers.Dense(11, activation='softmax')(dpl9)
output_digit2 = tf.keras.layers.Dense(11, activation='softmax')(dpl9)
output_digit3 = tf.keras.layers.Dense(11, activation='softmax')(dpl9)
output_digit4 = tf.keras.layers.Dense(11, activation='softmax')(dpl9)
output_digit5 = tf.keras.layers.Dense(11, activation='softmax')(dpl9)
# create model
model = tf.keras.models.Model(
input_layer, [output_digit1, output_digit2, output_digit3, output_digit4, output_digit5])
# optimizer
adam = tf.keras.optimizers.Adam(lr=0.00005)
model.compile(loss='categorical_crossentropy', optimizer=adam)
model.summary()
tf.keras.utils.plot_model(
model, to_file='svhnCNNModel.png', show_shapes=True)
return model
def train_model(model, X, y):
X = X/255.0
batch_size = 250
num_classes = 11
epochs = 200
save_dir = os.path.join(
os.getcwd(), '.SavedModels/multidigit')
model_name = 'svhn_keras_trained_model.h5'
y_label1 = y[:, 0, :]
y_label2 = y[:, 1, :]
y_label3 = y[:, 2, :]
y_label4 = y[:, 3, :]
y_label5 = y[:, 4, :]
y_labels = [y_label1, y_label2, y_label3, y_label4, y_label5]
early_stopper = tf.keras.callbacks.EarlyStopping(monitor='val_loss',
min_delta=0.001, patience=30,
restore_best_weights=True)
history = model.fit(X, y_labels, batch_size=batch_size,
epochs=epochs, validation_split=0.1, callbacks=[early_stopper])
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
model_path = os.path.join(save_dir, model_name)
model.save(model_path)
with open('./Data/saved_models/multidigit_history', 'wb') as f:
pickle.dump(history.history, f)
return model
def test_model(model, X, y):
X = X/255.0
y_label1 = y[:, 0, :]
y_label2 = y[:, 1, :]
y_label3 = y[:, 2, :]
y_label4 = y[:, 3, :]
y_label5 = y[:, 4, :]
y_labels = [y_label1, y_label2, y_label3, y_label4, y_label5]
scores = model.evaluate(X, y_labels)
print("Loss of model", scores[0])
print("Accuracy", scores[1])
predictions = model.predict(X)
return predictions, y
def main():
X_train, y_train, X_test, y_test = load_data(dataset='cropped')
model = create_model(X_train, y_train)
model = train_model(model, X_train, y_train)
scores, y = test_model(model, X_test, y_test)
return scores, y
if __name__ == "__main__":
main()