-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
199 lines (174 loc) · 7.27 KB
/
main.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
192
193
194
195
196
197
198
199
import os
import pickle
import shutil
import numpy as np
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.models import load_model
from keras.preprocessing import image
from sklearn.metrics import classification_report
import file_tools
import image_tools
import transfer_learning
from networks import MyLeNet, MyCNN, save_model_structure
def prepare_dataset(old, new, dirs_excluded):
print('[INFO] Preparing labels')
labels = file_tools.parse_labels(old, min_N=None, max_N=None,
exlude=dirs_excluded)
print(f'\t{len(labels)} classes in total')
print('[INFO] Preparing dataset')
os.makedirs(new, exist_ok=True)
file_tools.copy_dataset(old, new, labels)
print('[INFO] Dataset is ready!')
def fit_model(model_name, split_data, output_dir, save_metrics):
# Set values
max_epochs = 100
my_augmentation = False
keras_augmentation = True
min_train_size = 500
opt = 'adam'
shuffle_train = True
batch_size = 64
rows = 128
cols = 128
chans = 1
target_shape = (rows, cols, chans)
padding = 'ruling_gray'
if keras_augmentation:
keras_aug = image.ImageDataGenerator(
shear_range=0.1,
horizontal_flip=True,
vertical_flip=True,
rotation_range=30,
zoom_range=0.1,
width_shift_range=0.05,
height_shift_range=0.05,
fill_mode="nearest")
else:
keras_aug = None
# Extract image paths and labels
train_X, train_y, test_X, test_y, val_X, val_y, class_names = split_data
classes = len(class_names)
# Expand training set if necessary
if min_train_size > 0:
train_X, train_y = file_tools.broadcast_samples(train_X, train_y,
min_train_size)
# Prepare image generators
train_gen = image_tools.img_generator(
train_X, train_y, classes, batch_size,
target_shape, padding, shuffle_train, my_augmentation, keras_aug)
test_gen = image_tools.img_generator(
test_X, test_y, classes, batch_size, target_shape, padding)
val_gen = image_tools.img_generator(
val_X, val_y, classes, batch_size, target_shape, padding)
# for i in range(2):
# n = next(train_gen)[0][0]
# image_tools.plot_img(n)
print('[INFO] Preparing Model.')
if model_name == 'MyLeNet':
model = MyLeNet().build(rows, cols, chans, classes)
elif model_name == 'MyCNN':
model = MyCNN().build(rows, cols, chans, classes)
else:
raise ValueError(f'Unknown model name: {model_name}')
model.compile(loss='categorical_crossentropy', optimizer=opt,
metrics=['accuracy'])
# Stop training when validation loss hasn't decreased for 15 epochs
es = EarlyStopping(monitor='val_loss', mode='min', patience=5, verbose=1)
# Save only the best model based on validation accuracy
model_path = os.path.join(output_dir, 'best_model.h5')
mc = ModelCheckpoint(model_path, monitor='val_acc', mode='max',
save_best_only=True, verbose=1)
print('[INFO] Training Model.')
history = model.fit_generator(
train_gen,
steps_per_epoch=np.ceil(len(train_y)/batch_size),
validation_data=val_gen,
validation_steps=np.ceil(len(val_y)/batch_size),
epochs=max_epochs,
callbacks=[es, mc])
print('[INFO] Evaluating network.')
model = load_model(model_path)
predictions = model.predict_generator(
test_gen,
steps=np.ceil(len(test_y)/batch_size))
predictions = np.argmax(predictions, axis=1)
test_report = classification_report(test_y, predictions,
target_names=class_names)
print(test_report)
if save_metrics:
with open(os.path.join(output_dir, 'parameters.txt'), 'w') as fh:
fh.write(f'min_train_size: {min_train_size}\n')
fh.write(f'my_augmentation: {my_augmentation}\n')
fh.write(f'keras_augmentation: {keras_augmentation}\n')
fh.write(f'target_shape: {target_shape}\n')
fh.write(f'padding: {padding}\n')
fh.write(f'max_epochs: {max_epochs}\n')
fh.write(f'batch_size: {batch_size}\n')
fh.write(f'optimizer: {opt}')
with open(os.path.join(output_dir, 'test_report.txt'), 'w') as fh:
fh.write(test_report)
image_tools.plot_history(history, os.path.join(output_dir,
'training_progress.png'))
save_model_structure(model,
os.path.join(output_dir, 'network_summary.txt'))
print(f"[INFO] Model and metrics saved to '{output_dir}''")
def get_split_data(dataset, split_pickle=None, new=False):
if new:
split = (0.55, 0.25, 0.20)
min_N = 100
max_N = None
dist_file = 'output/class_distribution.txt'
name_file = 'output/class_names.txt'
train_paths, test_paths, val_paths = file_tools.train_test_split(
dataset, split, min_N, max_N, dist_file)
train_labels, test_labels, val_labels, class_names = \
file_tools.labels_from_paths(train_paths, test_paths, val_paths)
split_data = (train_paths, train_labels,
test_paths, test_labels,
val_paths, val_labels, class_names)
with open(split_pickle, 'wb') as fh:
pickle.dump(split_data, fh)
with open(name_file, 'w') as fh:
for name in class_names:
fh.write(f'{name}\n')
else:
with open(split_pickle, 'rb') as fh:
split_data = pickle.load(fh)
return split_data
def test_paths_to_images(split_data, test_dir_path):
test_X = split_data[2]
os.makedirs(test_dir_path, exist_ok=True)
for i in range(len(test_X)):
src = test_X[i]
label = os.path.basename(src)
dst = os.path.join(test_dir_path, label)
shutil.copy(src, dst)
def main():
"""Comment out parts that you don't want to run"""
# Dataset preparation (56843 images in final syke-dataset)
original_dataset = '/home/otso/Datasets/SYKE_150819'
new_dataset = '/home/otso/Datasets/syke'
dirs_excluded = ['summary', 'Unclassified']
prepare_dataset(original_dataset, new_dataset, dirs_excluded)
# Split image paths
dataset = '/home/otso/Datasets/syke'
split_data_path = 'output/split_data.pickle'
split_data = get_split_data(dataset, split_data_path, new=False)
# Copy test images to separate directory for testing purposes
test_paths_to_images(split_data, '/home/otso/Datasets/syke_test')
# Define model path
output_dir = 'models/test'
os.makedirs(output_dir, exist_ok=True)
save_metrics = True
# Train MyLeNet or MyCNN
fit_model('MyLeNet', split_data, output_dir, save_metrics)
# Extract features (transfer learning)
model_name = 'VGG19'
transfer_learning.extract_features(model_name, split_data,
output_dir, save_metrics)
# Train simple neural network on features
db_path = f'models/feat/{model_name}/features.h5'
transfer_learning.simple_feed_forward('SFF', db_path, output_dir, save_metrics)
print('[INFO] Finished!')
if __name__ == '__main__':
main()