-
Notifications
You must be signed in to change notification settings - Fork 5
/
train.py
336 lines (289 loc) · 15.2 KB
/
train.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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
# to get a cli tool
import click
# essentials
import numpy as np
import requests
import cv2
# For downloading and extracting dataset
from urllib.request import urlretrieve
import os
import tarfile
# for splitting into training & validation sets
from sklearn.model_selection import train_test_split
# for making labels one-hot encoded
from sklearn.preprocessing import LabelEncoder
from keras.utils import to_categorical
# for CNN and NN models
from keras.models import Sequential, Model
from keras.layers import Conv2D, Input, Dropout, Activation, Dense, MaxPooling2D, Flatten, GlobalAveragePooling2D
from keras.optimizers import Adadelta
from keras.callbacks import ModelCheckpoint
from keras.callbacks import EarlyStopping
# For transfer learning
from keras.applications.inception_v3 import InceptionV3
# for saving model
import json
# for pretty printing
import pprint
# for sorting dictionary by value
import operator
URL_CALTECH_101_DATA = 'http://www.vision.caltech.edu/Image_Datasets/Caltech101/101_ObjectCategories.tar.gz'
########################################
# Download & Extract dataset
########################################
def download_dataset(url):
current_directory = os.path.dirname(os.path.realpath('__file__'))
dataset_file_path = current_directory+"/dataset.tgz"
if os.path.exists(dataset_file_path):
print("Already downloaded.")
else:
filename, headers = urlretrieve(url, dataset_file_path)
print("Done")
def extract_dataset(dataset_file_path, extraction_directory):
if (not os.path.exists(extraction_directory)):
os.makedirs(extraction_directory)
if (dataset_file_path.endswith("tar.gz") or dataset_file_path.endswith(".tgz")):
tar = tarfile.open(dataset_file_path, "r:gz")
tar.extractall(path=extraction_directory)
tar.close()
elif (dataset_file_path.endswith("tar")):
tar = tarfile.open(dataset_file_path, "r:")
tar.extractall(path=extraction_directory)
tar.close()
print("Extraction of dataset done")
def get_current_directory():
return os.path.dirname(os.path.realpath(__file__))
########################################
# Creating Dataset
########################################
def get_images(object_category, data_directory):
if (not os.path.exists(data_directory)):
print("Data directory not found. Are you sure you downloaded and extracted dataset properly?")
return
obj_category_dir = os.path.join(os.path.join(data_directory,"101_ObjectCategories"),object_category)
images = [os.path.join(obj_category_dir,img) for img in os.listdir(obj_category_dir)]
return images
def read_image(image_path):
"""Read and resize individual images - Caltech 101 avg size of image is 300x200, so we resize accordingly"""
img = cv2.imread(image_path, cv2.IMREAD_COLOR)
img = cv2.resize(img, (300,200), interpolation=cv2.INTER_CUBIC)
return img
def return_images_per_category(data_directory):
categories = os.listdir(data_directory+"/101_ObjectCategories/")
object_images_count_dict = {}
for category in categories:
object_images_count_dict[category] = len(os.listdir(data_directory+"/101_ObjectCategories/"+category))
object_images_count_dict = sorted(object_images_count_dict.items(), key=operator.itemgetter(1), reverse=True)
return object_images_count_dict
def create_training_data(data_directory):
i = 0
X = np.ndarray((8677, 200, 300, 3), dtype=np.uint8)
Y = []
print("Preparing X and Y for dataset...")
for category,_ in return_images_per_category(data_directory):
if category == 'BACKGROUND_Google':
continue
print("Processing images of ",category)
for image in get_images(category, data_directory):
if not image.endswith('.jpg'):
# to escape hidden ipynb checkpoints and other unnecessary files
continue
X[i] = read_image(image)
Y.insert(i,category)
i += 1
print("Images processed : ",i+1," of 8678")
print("Datasets constructed")
return X,Y
def create_and_save_label_encoder(Y):
label_encoder = LabelEncoder()
Y_integer_encoded = label_encoder.fit_transform(Y)
np.save('label_encoder.npy', label_encoder.classes_)
return label_encoder, Y_integer_encoded
########################################
# Finding per category accuracy
########################################
def find_accuracy_per_category(data_directory, nn_model, label_encoder):
category_accuracy_dict = {}
for category,count in return_images_per_category(data_directory):
correctly_classified = 0
if category == 'BACKGROUND_Google':
continue
Y_category = [category for _ in range(count)]
encoded = label_encoder.transform(Y_category)
Y_category = to_categorical(encoded, num_classes=101)
assert Y_category.shape == (count, 101)
X_category = np.ndarray((count, 200, 300, 3), dtype=np.float64)
for i,img in enumerate(get_images(category, data_directory)):
if not img.endswith('.jpg'):
# to escape hidden ipynb checkpoints and other unnecessary files
continue
X_category[i] = read_image(img) / 255
score = nn_model.evaluate(x=X_category, y=Y_category, verbose=1)
del X_category
category_accuracy_dict[category] = score
print("Accuracy found for each class")
return category_accuracy_dict
def find_average_accuracy_for_model(data_directory, nn_model, label_encoder):
category_accuracy_dict = find_accuracy_per_category('./data', nn_model, label_encoder)
average_accuracy = 0
for category, scores in category_accuracy_dict.items():
print(category,":",scores[1])
average_accuracy += scores[1]
average_accuracy /= 101
print("Average accuracy : ",average_accuracy)
return average_accuracy, category_accuracy_dict
########################################
# COMMON STEPS
########################################
def perform_common_steps():
print("Downloading dataset...")
download_dataset(URL_CALTECH_101_DATA)
data_directory = os.path.join(get_current_directory(),'data')
extract_dataset(get_current_directory()+'/dataset.tgz',data_directory)
X, Y = create_training_data(data_directory)
label_encoder, Y_integer_encoded = create_and_save_label_encoder(Y)
X_normalized = X.astype(np.float64) / 255
del X
Y_one_hot = to_categorical(Y_integer_encoded)
X_train, X_validation, Y_train, Y_validation = train_test_split(X_normalized, Y_one_hot, test_size=0.25, random_state=42)
del X_normalized
return X_train, X_validation, Y_train, Y_validation, label_encoder, data_directory
########################################
# MODELS
########################################
def get_transfer_learning_model_architecture():
base_model = InceptionV3(weights='imagenet', include_top=False)
transfer_learning_arch = base_model.output
transfer_learning_arch = GlobalAveragePooling2D()(transfer_learning_arch)
transfer_learning_arch = Dense(1024, activation='relu')(transfer_learning_arch)
transfer_learning_arch = Dropout(0.4)(transfer_learning_arch)
transfer_learning_arch = Dense(512, activation='relu')(transfer_learning_arch)
transfer_learning_arch = Dropout(0.4)(transfer_learning_arch)
predictions = Dense(101, activation='softmax')(transfer_learning_arch)
transfer_learning_model = Model(inputs=base_model.input, outputs=predictions)
return transfer_learning_model
def get_simple_cnn_model_architecture():
model_cnn = Sequential()
model_cnn.add(Conv2D(16, (3,3), activation='relu', input_shape=(200,300,3)))
model_cnn.add(Conv2D(32, (3,3), activation='relu'))
model_cnn.add(MaxPooling2D(pool_size=2, strides=2))
model_cnn.add(Conv2D(64, (3,3), activation='relu'))
model_cnn.add(Conv2D(128, (3,3), activation='relu'))
model_cnn.add(MaxPooling2D(pool_size=2, strides=2))
model_cnn.add(Flatten())
model_cnn.add(Dense(101, activation='softmax'))
return model_cnn
def get_cnn_model_with_dropouts_architecture():
model_cnn_dropout = Sequential()
model_cnn_dropout.add(Conv2D(32, (3,3), activation='relu', input_shape=(200,300,3)))
model_cnn_dropout.add(Conv2D(32, (3,3), activation='relu'))
model_cnn_dropout.add(MaxPooling2D(pool_size=2, strides=2))
model_cnn_dropout.add(Conv2D(64, (3,3), activation='relu'))
model_cnn_dropout.add(Conv2D(64, (3,3), activation='relu'))
model_cnn_dropout.add(MaxPooling2D(pool_size=2, strides=2))
model_cnn_dropout.add(Flatten())
model_cnn_dropout.add(Dense(512, activation='relu'))
model_cnn_dropout.add(Dropout(0.5))
model_cnn_dropout.add(Dense(101, activation='softmax'))
return model_cnn_dropout
########################################
# COMMANDS
########################################
@click.group()
def cli():
""" This is a simple cli tool to train different neural network models for caltech 101 problem. """
pass
@cli.command()
@click.option('--print-average-accuracy/--no-print-average-accuracy', default=False, help='Depending upon which you '
'pass, model can print '
'average accuracy after it '
'has trained the model')
def train_transfer_learning_model(print_average_accuracy):
"""This trains the transfer learning model."""
X_train, X_validation, Y_train, Y_validation, label_encoder, data_directory = perform_common_steps()
# Get transfer learning model
transfer_learning_model = get_transfer_learning_model_architecture()
#We freeze the model excepted the added layers
#279 is number of mixed 9 layer
for layer in transfer_learning_model.layers[:280]:
layer.trainable = False
for layer in transfer_learning_model.layers[280:]:
layer.trainable = True
# Compile and fit the model
opt=Adadelta(lr=1.0, rho=0.9, epsilon=1e-08, decay=0.0)
transfer_learning_model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])
callbacks = [ModelCheckpoint('model_transfer_learning_weights.h5', monitor='val_acc', save_best_only=True),
EarlyStopping(monitor='val_loss', patience=4, verbose=1, mode='auto')]
transfer_learning_model.fit(X_train, Y_train, batch_size=32, epochs=15, verbose=1, validation_data=(X_validation,Y_validation), callbacks=callbacks)
print(">> Model trained.")
json_model = transfer_learning_model.to_json()
with open("model_transfer_learning.json", "w") as f:
json.dump(json.loads(json_model), f, indent=4)
print(">> Model saved to file.")
if print_average_accuracy:
print(">> Finding average accuracy per category")
total_avg_accuracy, category_accuracy_dict = find_average_accuracy_for_model(data_directory,
transfer_learning_model, label_encoder)
pp = pprint.PrettyPrinter(indent=2)
pp.pprint(category_accuracy_dict)
print(">> Total average accuracy across categories : ",str(total_avg_accuracy))
print("Done. Exiting.")
@cli.command()
@click.option('--print-average-accuracy/--no-print-average-accuracy', default=False, help='Depending upon which you '
'pass, model can print '
'average accuracy after it '
'has trained the model')
def train_simple_cnn_model(print_average_accuracy):
"""This trains a simple CNN model with 5 layers."""
X_train, X_validation, Y_train, Y_validation, label_encoder, data_directory = perform_common_steps()
# Get transfer learning model
model_cnn = get_simple_cnn_model_architecture()
# Compile and fit the model
model_cnn.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['accuracy'])
callbacks = [ModelCheckpoint('model_cnn_weights.h5', monitor='val_acc', save_best_only=True),
EarlyStopping(monitor='val_loss', patience=4, verbose=1, mode='auto')]
model_cnn.fit(X_train, Y_train, batch_size=64, epochs=30, verbose=1, validation_data=(X_validation,Y_validation), callbacks=callbacks)
print(">> Model trained.")
json_model = model_cnn.to_json()
with open("model_transfer_learning.json", "w") as f:
json.dump(json.loads(json_model), f, indent=4)
print(">> Model saved to file.")
if print_average_accuracy:
print(">> Finding average accuracy per category")
total_avg_accuracy, category_accuracy_dict = find_average_accuracy_for_model(data_directory,
model_cnn, label_encoder)
pp = pprint.PrettyPrinter(indent=2)
pp.pprint(category_accuracy_dict)
print(">> Total average accuracy across categories : ",str(total_avg_accuracy))
print("Done. Exiting.")
@cli.command()
@click.option('--print-average-accuracy/--no-print-average-accuracy', default=False, help='Depending upon which you '
'pass, model can print '
'average accuracy after it '
'has trained the model')
def train_cnn_model_with_dropouts(print_average_accuracy):
"""This trains an improvement over above simple CNN model which uses dropouts to reduce overfitting."""
X_train, X_validation, Y_train, Y_validation, label_encoder, data_directory = perform_common_steps()
# Get transfer learning model
model_cnn_dropout = get_cnn_model_with_dropouts_architecture()
# loss and optimizer
model_cnn_dropout.compile(loss='categorical_crossentropy',optimizer='sgd', metrics=['accuracy'])
# training
callbacks = [ModelCheckpoint('model_cnn_dropout_weights.h5', monitor='val_acc', save_best_only=True),
EarlyStopping(monitor='val_loss', patience=4, verbose=1, mode='auto')]
model_cnn_dropout.fit(X_train, Y_train, batch_size=16, epochs=10, verbose=1, validation_data=(X_validation,Y_validation), callbacks=callbacks)
print(">> Model trained.")
json_model = model_cnn_dropout.to_json()
with open("model_transfer_learning.json", "w") as f:
json.dump(json.loads(json_model), f, indent=4)
print(">> Model saved to file.")
if print_average_accuracy:
print(">> Finding average accuracy per category")
total_avg_accuracy, category_accuracy_dict = find_average_accuracy_for_model(data_directory,
model_cnn_dropout, label_encoder)
pp = pprint.PrettyPrinter(indent=2)
pp.pprint(category_accuracy_dict)
print(">> Total average accuracy across categories : ",str(total_avg_accuracy))
print("Done. Exiting.")
if __name__ == "__main__":
cli()