-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathvgg-retrain.py
159 lines (144 loc) · 6.19 KB
/
vgg-retrain.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
from keras import applications
from keras.preprocessing.image import ImageDataGenerator
from keras import optimizers
from keras.models import Sequential, Model
from keras.layers import Dropout, Flatten, Dense, GlobalAveragePooling2D
from keras import backend as k
from keras.callbacks import ModelCheckpoint, LearningRateScheduler, TensorBoard, EarlyStopping
from random import shuffle
from sklearn.metrics import accuracy_score, confusion_matrix
import matplotlib.pyplot as plt
import numpy as np
from sklearn.preprocessing import normalize
import itertools
import os
import shutil
import numpy as np
import seaborn as sns
import cv2
img_width, img_height = 120, 120
train_data_dir = "pictures/train"
validation_data_dir = "pictures/val"
test_data_dir = "pictures/test"
PIC_FOLDER = "pictures/processed_pictures"
batch_size = 16
epochs = 10
def split_dataset(clean=True):
if clean:
shutil.rmtree(train_data_dir)
shutil.rmtree(validation_data_dir)
shutil.rmtree(test_data_dir)
for dir_ in [train_data_dir, validation_data_dir, test_data_dir]:
if not os.path.exists(dir_):
os.mkdir(dir_)
n_train, n_val, n_test = 0, 0, 0
for dir_name, sub_dirs, file_list in os.walk(PIC_FOLDER):
for class_folder in sub_dirs:
curr_path = os.path.join(dir_name, class_folder)
imagepaths = [img for img in os.listdir(curr_path) if img != '.DS_Store']
shuffle(imagepaths)
n = len(imagepaths)
t1 = int(0.72*n)
t2 = int(0.8*n)
for train in imagepaths[:t1]:
dest = _create_dir_if_not_exists(train_data_dir, class_folder)
src = os.path.join(curr_path, train)
shutil.copy(src, dest)
n_train += 1
for val in imagepaths[t1:t2]:
dest = _create_dir_if_not_exists(validation_data_dir, class_folder)
src = os.path.join(curr_path, val)
shutil.copy(src, dest)
n_val += 1
for test in imagepaths[t2:]:
dest = _create_dir_if_not_exists(test_data_dir, class_folder)
src = os.path.join(curr_path, test)
shutil.copy(src, dest)
n_test += 1
return (n_train, n_val, n_test)
def _create_dir_if_not_exists(data_dir, class_folder):
fullpath = os.path.join(data_dir, class_folder)
if not os.path.exists(fullpath):
os.mkdir(fullpath)
return fullpath
def init_datagen(n_train, n_val, n_test):
train_datagen, val_datagen = _create_idg(True), _create_idg()
train_generator = train_datagen.flow_from_directory(train_data_dir, target_size=(img_height, img_width), batch_size=batch_size, class_mode='categorical')
val_generator = val_datagen.flow_from_directory(validation_data_dir, target_size=(img_height, img_width), batch_size=batch_size, class_mode='categorical')
return (train_generator, val_generator)
def train_model(train_generator, val_generator):
model = _init_model()
checkpoint = ModelCheckpoint('vgg16_1.h5', monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=False, mode='auto', period=1)
early = EarlyStopping(monitor='val_acc', min_delta=0, patience=5, verbose=1, mode='auto')
model.fit_generator(train_generator,
samples_per_epoch=train_generator.n,
epochs=epochs,
validation_data=val_generator,
nb_val_samples=val_generator.n,
callbacks=[checkpoint, early])
return model
def _init_model():
model = applications.VGG19(weights = 'imagenet', include_top=False, input_shape=(img_width, img_height, 3))
x = model.output
x = Flatten()(x)
x = Dense(1024, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(1024, activation='relu')(x)
preds = Dense(8, activation='softmax')(x)
model_final = Model(input=model.input, output=preds)
model_final.compile(loss='categorical_crossentropy', optimizer=optimizers.SGD(lr=0.0001, momentum=0.9), metrics=['accuracy'])
return model_final
def _create_idg(train=False):
return ImageDataGenerator(horizontal_flip=True,
fill_mode='nearest',
zoom_range=0.3,
width_shift_range=0.3,
height_shift_range=0.3,
rotation_range=30)
def create_test():
cnt = 0
X, y = [], []
for classname in os.listdir(test_data_dir):
if classname != '.DS_Store':
cnt += 1
currpath = os.path.join(test_data_dir, classname)
for img in os.listdir(currpath):
imagepath = os.path.join(currpath, img)
x = cv2.imread(imagepath)
X.append(x)
y.append(cnt)
return np.array(X), np.array(y)
def evaluate_model(model, X_train, y_true):
y_pred = model.predict(X_train)
assert len(y_pred) == len(y_true)
acc = accuracy_score(y_true, y_pred)
print("Test Accuracy: {:.4f}".format(acc))
cm = confusion_matrix(y_true, y_pred)
classes = [f for f in os.listdir(PIC_FOLDER) if f != '.DS_Store']
_create_cm(cm, classes)
def _create_cm(cm, classes, save_dest=None):
plt.imshow(cm, interpolation='nearest', cmap=plt.cm.Blues)
plt.title("Test Set Confusion Matrix")
plt.colorbar()
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
fmt = '.2f' if normalize else 'd'
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, format(cm[i, j], fmt),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
if save_dest:
plt.savefig(save_dest)
else:
plt.show()
if __name__=="__main__":
n_train, n_val, n_test = split_dataset(clean=False)
train_generator, val_generator = init_datagen(n_train, n_val, n_test)
fitted_model = train_model(train_generator, val_generator)
X_train, y_true = create_test()
evaluate_model(fitted_model, X_train, y_true)