def buildModel(self, dense_list):
'''
n_filters: num of filters in the initial layers
depth: depth of the resnet
For dense layers, the dropouts are always 0.5
den_list: key arg list for den layers
'''
in_layer = Input(self.input_shape)
# use keras existing resnet50
xceptionModel = Xception(include_top=False, weights='imagenet',
input_tensor=Input(shape=self.input_shape), pooling="avg")
kernel = Dropout(0.5) (xceptionModel (in_layer))
xceptionModel.summary()
print kernel.shape
denlayer = kernel
# denlayer = GlobalMaxPooling2D() (kernel)
# denlayer = Flatten() (kernel)
# adding dense layers
for kargs in dense_list:
denlayer = Dropout(0.5) (Dense(**kargs) (denlayer))
out_layer = Dense(self.output_dim, activation='softmax') (denlayer)
self.model = Model(inputs=[in_layer], outputs=[out_layer])
def load_xcep_model(img_batch):
model = Xception(include_top=True,
weights=MODEL_PATH,
input_shape=(input_height, input_width, 3),
classes=classes)
model.summary()
plot_model(model, 'xcep_model.jpg')
# preds = model.predict(img_batch)
return model, img_batch
def extract_features(directory):
# load the model
model = Xception()
# re-structure the model
model.layers.pop()
model = Model(inputs=model.inputs, outputs=model.layers[-1].output)
# summarize
print(model.summary())
# extract features from each photo
features = dict()
for name in listdir(directory):
# load an image from file
filename = directory + '/' + name
image = load_img(filename, target_size=(299, 299))
# convert the image pixels to a numpy array
image = img_to_array(image)
# reshape data for the model
image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
# prepare the image for the VGG model
image = preprocess_input(image)
# get features
feature = model.predict(image, verbose=0)
# get image id
image_id = name.split('.')[0]
# store feature
features[image_id] = feature
print('>%s' % name)
return features
class Xceptionmodel(CNNmodel):
def __init__(self, data_directory, batch_size=16):
self.cnn = CNNmodel(data_directory)
self.height, self.width, self.depth = (48, 48, 1)
self.batch_size = batch_size
self.xception = Xception(include_top=True,
weights=None,
input_shape=(self.height, self.width,
self.depth),
classes=4)
def summary(self):
print(self.xception.summary())
def train(self,
optimizer='adam',
loss='categorical_crossentropy',
epoch=20):
self.xception.compile(loss="categorical_crossentropy",
optimizer=optimizer,
metrics=["accuracy"])
print("[INFO] training network...")
es = EarlyStopping(monitor='val_acc',
mode='max',
verbose=1,
patience=4)
self.history = self.xception.fit(self.cnn.train_data,
epochs=epoch,
validation_data=self.cnn.val_data,
verbose=1,
callbacks=[es])
print("[INFO] Saving model...")
pickle.dump(self.xception, open('xception.pkl', 'wb'))
print("[INFO] Done...")
from keras.applications import VGG16 from keras.applications import VGG19 from keras.applications import Xception from keras.applications import ResNet50 from keras.applications import MobileNet from keras.applications import MobileNetV2 from keras.applications import InceptionV3 from keras.applications import InceptionResNetV2 # conv_base = VGG16(weights='imagenet', include_top=False, # input_shape=(150,150,3)) # conv_base = VGG16() # 224,224,3 conv_base = Xception() conv_base.summary() from keras import models, layers model = models.Sequential() model.add(conv_base) # model.add(layers.Flatten()) model.add(layers.Dense(256, activation='relu')) model.add(layers.Dense(1, activation='sigmoid')) model.summary()
## define the model (preset weights)
print('[INFO] defining model...')
if worker == 'single':
model = Xception(include_top=True,
weights=None,
classes=len(labeltonumber))
elif worker == 'parallel':
with tf.device('/cpu:0'):
model = Xception(include_top=True,
weights=None,
classes=len(labeltonumber))
## print a summary of the model
print(model.summary())
## values from Olafenwa and Olafenva - 2018 and
## https://machinelearningmastery.com/evaluate-performance-deep-learning-models-keras/
EPOCHS = 200
print('epochs: {}'.format(EPOCHS))
## batch normalization batch sizes: 64, 128, 256, 512
BATCHSIZE = 32
print('batchsize: {}'.format(BATCHSIZE))
STEPS_PER_EPOCH = len(X_train) / BATCHSIZE
print('steps per epoch: {}'.format(STEPS_PER_EPOCH))
## Olafenwa and Olafenva - 2018 #######################
## step decay
def lr_schedule(epoch):
from keras.applications import VGG16, ResNet50, InceptionV3, Xception, DenseNet121 from keras.utils import plot_model # https://github.com/keras-team/keras-applications/tree/master/keras_applications vgg16 = VGG16(weights=None) vgg16.summary() plot_model(vgg16, "vgg16.png", show_layer_names=True, show_shapes=True) resnet50 = ResNet50(weights=None) resnet50.summary() plot_model(resnet50, "resnet50.png", show_layer_names=True, show_shapes=True) inception = InceptionV3(weights=None) inception.summary() plot_model(inception, "inception.png", show_layer_names=True, show_shapes=True) xception = Xception(weights=None) xception.summary() plot_model(xception, "xception.png", show_layer_names=True, show_shapes=True) densenet = DenseNet121(weights=None) densenet.summary() plot_model(densenet, "densenet.png", show_layer_names=True, show_shapes=True)
def Xception_Net(trainable=None):
# Preprocessing the dataset into keras feedable format
train_datagen = ImageDataGenerator(
rotation_range = rotation,
width_shift_range = width_shift,
height_shift_range= height_shift,
rescale= scale,
shear_range= shear,
zoom_range= zoom,
horizontal_flip= horizontal,
fill_mode=fill,
validation_split=validation
)
test_datagen = ImageDataGenerator(
rescale= scale,
)
train_generator = train_datagen.flow_from_directory(
path,
target_size=target,
batch_size=batch,
class_mode='categorical',
subset='training',
)
validation_generator = train_datagen.flow_from_directory(
path,
target_size=target,
batch_size=batch,
class_mode='categorical',
subset='validation'
)
# Loading the ResNet50 Model
xception = Xception(include_top=False, weights='imagenet', input_shape=input_sh)
output = xception.layers[-1].output
output = keras.layers.Flatten()(output)
xception = Model(xception.input, output=output)
print(xception.summary())
print('\n\n\n')
# If you chose not for fine tuning
if trainable is None:
model = Sequential()
model.add(xception)
model.add(Dense(hidden, activation='relu', input_dim=input_sh))
model.add(Dropout(dropout_num))
model.add(Dense(hidden, activation='relu'))
model.add(Dropout(dropout_num ))
if classes == 1:
model.add(Dense(classes, activation='sigmoid', name='Output'))
else:
model.add(Dense(classes, activation='softmax', name='Output'))
for layer in xception.layers:
layer.trainable = False
print("The model summary of Xception -->\n\n\n") # In this the Resnet50 layers are not trainable
for i, layer in enumerate(xception.layers):
print(i, layer.name, layer.trainable)
model.compile(loss=loss_param, # Change according to data
optimizer=optimizers.RMSprop(),
metrics=['accuracy'])
print("The summary of final Model \n\n\n")
print(model.summary())
print('\n\n\n')
fit_history = model.fit_generator(
train_generator,
steps_per_epoch=len(train_generator.filenames) // batch,
epochs=epoch,
shuffle=True,
validation_data=validation_generator,
validation_steps=len(train_generator.filenames) // batch,
class_weight=n,
callbacks=[
EarlyStopping(patience=patience_param, restore_best_weights=True),
ReduceLROnPlateau(patience=patience_param)
])
os.chdir(output_path)
model.save("model.h5")
print(fit_history.history.keys())
plt.figure(1, figsize = (15,8))
plt.subplot(221)
plt.plot(fit_history.history['accuracy'])
plt.plot(fit_history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'valid'])
plt.subplot(222)
plt.plot(fit_history.history['loss'])
plt.plot(fit_history.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'valid'])
plt.show()
if trainable is not None:
# Make last block of the conv_base trainable:
for layer in xception.layers[:trainable]:
layer.trainable = False
for layer in xception.layers[trainable:]:
layer.trainable = True
print('Last block of the conv_base is now trainable')
for i, layer in enumerate(xception.layers):
print(i, layer.name, layer.trainable)
model = Sequential()
model.add(xception)
model.add(Dense(hidden, activation='relu', input_dim=input_sh))
model.add(Dropout(dropout_num))
model.add(Dense(hidden, activation='relu'))
model.add(Dropout(dropout_num ))
model.add(Dense(hidden, activation='relu'))
model.add(Dropout(dropout_num ))
if classes == 1:
model.add(Dense(classes, activation='sigmoid', name='Output'))
else:
model.add(Dense(classes, activation='softmax', name='Output'))
for layer in xception.layers:
layer.trainable = False
print("The model summary of Xception -->\n\n\n") # In this the Resnet50 layers are not trainable
model.compile(loss=loss_param, # Change according to data
optimizer=optimizers.RMSprop(),
metrics=['accuracy'])
print("The summary of final Model \n\n\n")
print(model.summary())
print('\n\n\n')
fit_history = model.fit_generator(
train_generator,
steps_per_epoch=len(train_generator.filenames) // batch,
epochs=epoch,
shuffle=True,
validation_data=validation_generator,
validation_steps=len(train_generator.filenames) // batch,
class_weight=n,
callbacks=[
EarlyStopping(patience=patience_param, restore_best_weights=True),
ReduceLROnPlateau(patience=patience_param)
])
os.chdir(output_path)
model.save("model.h5")
print(fit_history.history.keys())
plt.figure(1, figsize = (15,8))
plt.subplot(221)
plt.plot(fit_history.history['accuracy'])
plt.plot(fit_history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'valid'])
plt.subplot(222)
plt.plot(fit_history.history['loss'])
plt.plot(fit_history.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'valid'])
plt.show()
def get_features_CNN(image_list, method='VGG16'):
"""
Extracts image features using CNN
Arguments:
- image_list: list, image set
- method: str, VGG16 or Xception
Returns:
- features: list, contains tuples with image path + histogram of features
"""
features = []
list_features_batch = []
if method in ['VGG16', 'VGG16DAB']:
print('Loading network...')
model = VGG16(weights='imagenet', include_top=False, pooling='avg')
model.summary()
x = 0
n = 0
for im in tqdm(image_list):
image = imread(im)
if method == 'VGG16DAB':
image = imagetoDAB(image)
image = numpy.asarray(image)
image = numpy.expand_dims(image, axis=0)
image = imagenet_utils.preprocess_input(image)
curr_feat = model.predict(image)
curr_feat = curr_feat.flatten()
features.append((im, curr_feat))
n += 1
# When list has 100000 images + features, it is saved with pickle and a new list starts
if n == 100000:
features_batch = os.path.join(
outpath, 'features_{}_batch{}.p'.format(method, x))
with open(features_batch, "wb") as f:
pickle.dump(features, f)
list_features_batch.append(features_batch)
features = []
n = 0
x += 1
features_batch = os.path.join(
outpath, 'features_{}_batch{}.p'.format(method, x))
with open(features_batch, "wb") as f:
pickle.dump(features, f)
list_features_batch.append(features_batch)
if method in ['Xception', 'XceptionDAB']:
print('Loading network...')
model = Xception(weights='imagenet', include_top=False, pooling='avg')
model.summary()
x = 0
n = 0
for im in tqdm(image_list):
image = imread(im)
if method == 'XceptionDAB':
image = imagetoDAB(image)
image = numpy.asarray(image)
image = numpy.expand_dims(image, axis=0)
image = preprocess_input(image)
curr_feat = model.predict(image)
curr_feat = curr_feat.flatten()
features.append((im, curr_feat))
n += 1
# When list has 100000 images + features, it is saved with pickle and a new list starts
if n == 100000:
features_batch = os.path.join(
outpath, 'features_{}_batch{}.p'.format(method, x))
with open(features_batch, "wb") as f:
pickle.dump(features, f)
list_features_batch.append(features_batch)
features = []
n = 0
x += 1
features_batch = os.path.join(
outpath, 'features_{}_batch{}.p'.format(method, x))
with open(features_batch, "wb") as f:
pickle.dump(features, f)
list_features_batch.append(features_batch)
return list_features_batch
width),
batch_size=batch_size,
seed=seed,
class_mode='categorical')
# Test generator
test_datagen = ImageDataGenerator(rescale=1. / 255)
test_generator = test_datagen.flow_from_directory(test_dir,
target_size=(height, width),
batch_size=batch_size,
seed=seed,
class_mode='categorical')
base_model = Xception(weights=weights,
include_top=False,
input_shape=(height, width, channels))
base_model.summary()
def extract_features(sample_count, datagen):
start = dt.datetime.now()
features = np.zeros(shape=(sample_count, 5, 5, 2048))
labels = np.zeros(shape=(sample_count, 10))
generator = datagen
i = 0
for inputs_batch, labels_batch in generator:
stop = dt.datetime.now()
time = (stop - start).seconds
print('\r',
'Extracting features from batch',
str(i + 1),
'/',
model = DenseNet121()
model = DenseNet169()
model = DenseNet201()
model = NASNetLarge()
model = NASNetMobile()
vgg16 = VGG16(
weights=None,
include_top=False,
classes=10,
input_shape=(224, 224, 3),
# classifier_activation="softmax"
)
# (None, 224, 224, 3)
# 모델 다운 받아 와야함
vgg16.summary()
# 잘만든 모델 가져다 쓰는 거 = 전이학습
# 이미지 모델에서 준우승 한 모델
# VGG16이랑 엮기
model = Sequential()
model.add(vgg16)
model.add(Dense(256))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dense(10, activation='softmax'))
model.summary()
from keras.layers import Dense, Flatten, Dropout
from keras.optimizers import RMSprop
from keras.models import Sequential
from keras.preprocessing.image import ImageDataGenerator
base_dir = "D:/dataset/images/kaggle/"
train_dir = os.path.join(base_dir, "train")
test_dir = os.path.join(base_dir, 'test')
validation_dir = os.path.join(base_dir, 'validation')
num_classes = 12
import matplotlib.pyplot as plt
conv_base = Xception(include_top=False,
weights='imagenet',
input_shape=(71, 71, 3))
print(conv_base.summary())
datagen = ImageDataGenerator(rescale=1. / 255,
rotation_range=0.2,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest')
train_gen = datagen.flow_from_directory(train_dir,
target_size=(71, 71),
batch_size=20,
class_mode='categorical')
val_gen = test_datagen.flow_from_directory(
val_dir,
target_size=(299,299),
batch_size=20,
class_mode='binary')
test_gen = test_datagen.flow_from_directory(
test_dir,
target_size=(299,299),
batch_size=24,
class_mode='binary')
#Xception model
conv_xception = Xception(include_top=False, weights='imagenet', input_shape = (299, 299, 3))
conv_xception.summary()
def XceptionModel():
conv_xception.trainable = False
# Model create
model = models.Sequential()
model.add(conv_xception)
model.add(layers.Flatten())
model.add(layers.BatchNormalization())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation = 'relu'))
model.add(layers.Dense(1, activation = 'sigmoid'))
print(model.summary())
#model compile
opt = optimizers.Adam(lr = 1e-4)
model.compile(loss = 'binary_crossentropy', optimizer = opt, metrics = ['acc'])
return model
