def transfer_model(model_name, input_shape, classes_nr):
new_input = Input(shape=(input_shape[0], input_shape[1], 3))
if model_name == "vgg16":
model = VGG16(include_top=False, input_tensor=new_input)
if model_name == "densenet121":
model = DenseNet121(include_top=False, input_tensor=new_input)
if model_name == "inceptionv3":
model = InceptionV3(include_top=False, input_tensor=new_input)
if model_name == "mobilenet":
model = MobileNet(include_top=False, input_tensor=new_input)
if model_name == "resnet101":
model = ResNet101(include_top=False, input_tensor=new_input)
if model_name == "xception":
model = Xception(include_top=False, input_tensor=new_input)
for layer in model.layers:
layer.trainable = False
flat1 = layers.Flatten()(model.layers[-1].output)
class1 = layers.Dense(1024, activation='relu')(flat1)
drop1 = layers.Dropout(0.2)(class1)
class2 = layers.Dense(256, activation='relu')(drop1)
output = layers.Dense(classes_nr, activation='softmax')(class2)
model = Model(inputs=model.inputs, outputs=output)
return model
def __init__(self, data_format='channels_last'):
"""Constructor function."""
# determine the input shape
if data_format == 'channels_last':
input_shape = (IMAGE_SIZE[0], IMAGE_SIZE[1], 3)
elif data_format == 'channels_first':
input_shape = (3, IMAGE_SIZE[0], IMAGE_SIZE[1])
else:
raise ValueError('unrecognized data format: ' + data_format)
# build the Inception-v3 backbone & final classification layer
K.set_image_data_format(data_format)
import ssl
ssl._create_default_https_context = ssl._create_unverified_context
# net = InceptionV3()
# print("==============",net)
net = InceptionV3(include_top=False,
weights=None,
input_tensor=None,
input_shape=input_shape,
backend=K)
x = net.output
x = Conv2D(NUM_CLASSES,
8,
data_format=data_format,
activation='softmax',
name='output')(x)
x = Flatten(data_format=data_format)(x)
net_final = Model(inputs=net.input, outputs=x)
# obtain input & output tensors
self.inputs = net_final.inputs
self.outputs = net_final.outputs
def main():
from tensorflow.python.keras.models import Model
from tensorflow.python.keras.layers import Flatten, Dense, Dropout, Conv2D
from tensorflow.python.keras.applications.inception_v3 import InceptionV3, preprocess_input
data_format = 'channels_last'
input_shape = (299, 299, 3) if data_format == 'channels_last' else (3, 299,
299)
#input_shape = (2048, 2048, 3) if data_format == 'channels_last' else (3, 2048, 2048)
K.set_image_data_format(data_format)
net = InceptionV3(include_top=False,
weights=None,
input_tensor=None,
input_shape=input_shape)
x = net.output
x = Conv2D(2, 8, activation='softmax', name='output')(x)
model = Model(inputs=net.input, outputs=x)
model.load_weights(h5_path, by_name=True)
converted_output_node_names = [node.op.name for node in model.outputs]
print(('Converted output node names are: %s',
str(converted_output_node_names)))
sess = K.get_session()
constant_graph = graph_util.convert_variables_to_constants(
sess, sess.graph.as_graph_def(), converted_output_node_names)
graph_io.write_graph(constant_graph, save_path, 'model.pb', as_text=False)
sess.close()
def InceptionV3WithCustomLayers(nb_classes, input_shape, fc_size):
"""
Adding custom final layers on Inception_V3 model with imagenet weights
Args:
nb_classes: # of classes
input_shape: input shape of the images
Returns:
new keras model with new added last layer/s and the base model which new layers are added
"""
base_model = InceptionV3(input_tensor=Input(shape=input_shape),
weights='imagenet',
include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(fc_size * 2, activation='relu')(x) # new FC layer, random init
x = Dropout(0.3)(x)
# x = Dense(FC_SIZE, activation='relu')(x) # new FC layer, random init
# x = Dropout(0.5)(x)
# x = Dense(FC_SIZE * 4, activation='relu')(x) # new FC layer, random init
# x = Dropout(0.5)(x)
predictions = Dense(nb_classes,
activation='softmax')(x) # new softmax layer
model = Model(outputs=predictions, inputs=base_model.input)
return model, base_model
def load_model(model_arch):
if model_arch == 'inceptionV3':
model = InceptionV3(include_top=False,
weights='imagenet',
pooling='avg')
height, width = 299, 299
else:
raise NotImplementedError("That architecture is not implemented yet")
return model, (height, width)
def create_model(num_hidden, num_classes):
base_model = InceptionV3(include_top=False, weights='imagenet')
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(num_hidden, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)
for layer in base_model.layers:
layer.trainable = False
model = Model(inputs=base_model.input, outputs=predictions)
return model
base_model = InceptionV3(include_top=False, weights='image')
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(num_hidden, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)
for layer in base_model.layers:
layer.trainable = False
model = Model(inputs=base_model.input, outputs=predictions)
return model
def get_model():
_model = InceptionV3(input_shape=(78, 78, 3),
include_top=False,
weights=None)
_model.load_weights(weights_file)
for layer in _model.layers:
layer.trainable = False
_model.summary()
return _model
def setUp(self):
self.model = InceptionV3(input_tensor=Input(shape=utils.INPUT_SHAPE),
weights=None,
include_top=True,
classes=3)
self.layers_bool = []
bool = True
for layer in self.model.layers:
layer.trainable = bool
self.layers_bool.append(layer.trainable)
# switch true and false every iteration
bool = not bool
def classification(
shape: Tuple[int, int, int] = (224, 224, 3),
n_class: int = 3,
model_net: str = "Resnet50V2",
resnet_train: bool = True,
) -> Model:
"""
Modelo de classificação entre covid, normal e pneumonia
Args:
-----
input_size (tuple, optional): Tamanho da imagem de entrada.
Defaults to (224, 224, 3).
n_class (int, optional): Número de classes de saída.
Defaults to 3.
Returns:
--------
(keras.Model) : Modelo do keras
"""
inputs = Input(shape, name="entrada_modelo")
model = Conv2D(
filters=3,
kernel_size=(3, 3),
padding="same",
activation="relu",
name="conv_gray_rgb",
)(inputs)
params = {
"include_top": False,
"weights": "imagenet",
"input_shape": (shape[0], shape[1], 3),
"pooling": "avg",
}
if model_net == "VGG19":
base_model = VGG19(**params)
elif model_net == "InceptionResNetV2":
base_model = InceptionV3(**params)
elif model_net == "MobileNetV2":
base_model = MobileNetV3Small(**params)
elif model_net == "DenseNet201":
base_model = DenseNet201(**params)
else:
base_model = ResNet50V2(**params)
base_model.trainable = resnet_train
model = base_model(model)
model = Dropout(0.5, name="drop_0")(model)
model = Dense(units=256, name="dense_0")(model)
model = Dropout(0.5, name="drop_1")(model)
model = Dense(units=n_class, name="classifier")(model)
predictions = Activation(activation="softmax", name="output")(model)
return Model(inputs=inputs, outputs=predictions)
def define_model(IMAGE_DIMS, VEC_LEN, weight_dir):
i10 = Input(shape=IMAGE_DIMS)
i20 = Input(shape=IMAGE_DIMS)
i30 = Input(shape=IMAGE_DIMS)
t1 = Input(shape=(1, ))
t2 = Input(shape=(1, ))
print("[INFO] Weights restored from pre-trained InceptionV3!")
encoder = InceptionV3(weights=weight_dir, include_top=False)
pooling = GlobalAveragePooling2D()
def l2_norm_(x):
return K.sqrt(K.sum(K.square(x), 1))
def l2_normalize(x):
return K.l2_normalize(x, 1)
output = Dense(VEC_LEN, activation='sigmoid', name='encoder_output')
o1 = encoder(i10)
o2 = encoder(i20)
o3 = encoder(i30)
o1 = pooling(o1) # 全局平均池化层
o1 = output(o1) # 有1024个节点的全连接层
o2 = pooling(o2) # 全局平均池化层
o2 = output(o2) # 有1024个节点的全连接层
o3 = pooling(o3) # 全局平均池化层
o3 = output(o3) # 有1024个节点的全连接层
def distance(inputs):
ap, an, margin, gthr = inputs
ap_l2n = K.sqrt(K.sum(K.square(ap), axis=1, keepdims=True))
an_l2n = K.sqrt(K.sum(K.square(an), axis=1, keepdims=True))
d = K.minimum((an_l2n - ap_l2n), margin)
g = K.maximum(ap_l2n, gthr)
y = K.concatenate([d, g], axis=1)
return y
ap = Subtract()([o1, o2])
an = Subtract()([o1, o3])
val = Lambda(distance, name='margin')([ap, an, t1, t2])
model = Model(inputs=[i10, i20, i30, t1, t2], outputs=val)
return model
def create_model(self):
'''
InceptionV3を用いた機械学習モデルを構築します。
戻り値：
compile済みモデルオブジェクト
'''
# inputの定義：cam/img_array 形式
img_in = Input(shape=self.image_shape, name='img_in')
# InceptionV3 canned model を使用
x = InceptionV3(weights='imagenet', include_top=False)(img_in)
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='softmax')(x)
x = Dropout(0.5)(x)
# 従来モデルより流用
x = Flatten(name='flattened')(x)
x = Dense(100, activation='relu')(x)
x = Dropout(.1)(x)
x = Dense(50, activation='relu')(x)
x = Dropout(.1)(x)
# outputs[0] の定義：15分類確率ベクトル
angle_out = Dense(15, activation='softmax', name='angle_out')(x)
# outputs[1] の定義：スロットルPWM値
throttle_out = Dense(1, activation='relu', name='throttle_out')(x)
print('model throttle_out:' + str(throttle_out))
# モデルオブジェクトの定義
model = Model(inputs=[img_in], outputs=[angle_out, throttle_out])
model.compile(optimizer='adam',
loss={
'angle_out': 'categorical_crossentropy',
'throttle_out': 'mean_absolute_error'
},
loss_weights={
'angle_out': 0.9,
'throttle_out': .001
})
return model
def build_baseline_model(args, input_shape):
"""
Builds a baseline InceptionV3 model from tensorflow implementation
with no trained weights loaded and including top layers for prediction
Args:
args: necessary args needed for training like train_data_dir, batch_size etc...
input_shape: shape of input tensor
Returns:
baseline inceptionV3 model
"""
iv3 = InceptionV3(input_tensor=Input(shape=input_shape),
weights=None,
include_top=True,
classes=args.nb_classes)
iv3.compile(optimizer='RMSprop',
loss='categorical_crossentropy',
metrics=['accuracy'])
return iv3
def get_model(self, model_name, weights):
if model_name == 'InceptionV3':
from tensorflow.python.keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights=weights, include_top=False)
elif model_name == 'NASNetLarge':
from tensorflow.python.keras.applications.nasnet import NASNetLarge
base_model = NASNetLarge(weights=weights, include_top=False)
elif model_name == 'DenseNet201':
from tensorflow.python.keras.applications.densenet import DenseNet201
base_model = DenseNet201(weights=weights, include_top=False)
elif model_name == 'Xception':
from tensorflow.python.keras.applications.xception import Xception
base_model = Xception(weights=weights, include_top=False)
elif model_name == 'VGG16':
from tensorflow.python.keras.applications.vgg16 import VGG16
base_model = VGG16(weights=weights, include_top=False)
elif model_name == 'VGG19':
from tensorflow.python.keras.applications.vgg19 import VGG19
base_model = VGG19(weights=weights, include_top=False)
elif model_name == 'NASNetMobile':
from tensorflow.python.keras.applications.nasnet import NASNetMobile
base_model = NASNetMobile(weights=weights, include_top=False)
elif model_name == 'MobileNetV2':
from tensorflow.python.keras.applications.mobilenet_v2 import MobileNetV2
base_model = MobileNetV2(weights=weights, include_top=False)
elif model_name == 'ResNet50':
from tensorflow.python.keras.applications.resnet50 import ResNet50
base_model = ResNet50(weights=weights, include_top=False)
elif model_name == 'InceptionResNetV2':
from tensorflow.python.keras.applications.inception_resnet_v2 import InceptionResNetV2
base_model = InceptionResNetV2(weights=weights, include_top=False, )
else:
raise KeyError('Unknown network.')
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(512, activation='relu')(x)
predictions = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=predictions)
return model
def get_model(self, model_name, weights='imagenet'):
if model_name == 'InceptionV3':
from tensorflow.python.keras.applications.inception_v3 import InceptionV3
base_model = InceptionV3(weights=weights, include_top=False)
elif model_name == 'NASNetLarge':
from tensorflow.python.keras.applications.nasnet import NASNetLarge
base_model = NASNetLarge(weights=weights, include_top=False)
elif model_name == 'DenseNet201':
from tensorflow.python.keras.applications.densenet import DenseNet201
base_model = DenseNet201(weights=weights, include_top=False)
elif model_name == 'Xception':
from tensorflow.python.keras.applications.xception import Xception
base_model = Xception(weights=weights, include_top=False)
elif model_name == 'VGG19':
from tensorflow.python.keras.applications.vgg19 import VGG19
base_model = VGG19(weights=weights, include_top=False)
elif model_name == 'NASNetMobile':
from tensorflow.python.keras.applications.nasnet import NASNetMobile
base_model = NASNetMobile(weights=weights, include_top=False)
elif model_name == 'MobileNetV2':
from tensorflow.python.keras.applications.mobilenet_v2 import MobileNetV2
base_model = MobileNetV2(weights=weights, include_top=False)
elif model_name == 'ResNet50':
from tensorflow.python.keras.applications.resnet50 import ResNet50
base_model = ResNet50(weights=weights, include_top=False)
elif model_name == 'InceptionResNetV2':
from tensorflow.python.keras.applications.inception_resnet_v2 import InceptionResNetV2
base_model = InceptionResNetV2(weights=weights, include_top=False)
else:
raise KeyError('Unknown network.')
x = base_model.output
x = GlobalAveragePooling2D()(x)
# x = Dense(1024, activation='relu')(x)
output = Dense(1, activation='sigmoid')(x)
# output = SiameseLossLayer(self.batch_size, self.num_distort, self.num_level)(x)
model = Model(inputs=base_model.input, outputs=output)
self.name = model_name
return model
def train(args):
"""Use transfer learning and fine-tuning to train a network on a new dataset"""
nb_train_samples = get_nb_files(args.train_dir)
nb_classes = len(glob.glob(args.train_dir + "/*"))
nb_val_samples = get_nb_files(args.val_dir)
nb_epoch = int(args.nb_epoch)
batch_size = int(args.batch_size)
# data prep
train_datagen = ImageDataGenerator(preprocessing_function=preprocess_input,
rotation_range=30,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
test_datagen = ImageDataGenerator(preprocessing_function=preprocess_input,
rotation_range=30,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
train_generator = train_datagen.flow_from_directory(
args.train_dir,
target_size=(IM_WIDTH, IM_HEIGHT),
batch_size=batch_size,
)
validation_generator = test_datagen.flow_from_directory(
args.val_dir,
target_size=(IM_WIDTH, IM_HEIGHT),
batch_size=batch_size,
)
# setup model
base_model = InceptionV3(
weights='imagenet',
include_top=False) #include_top=False excludes final FC layer
model = add_new_last_layer(base_model, nb_classes)
# transfer learning
setup_to_transfer_learn(model, base_model)
history_tl = model.fit_generator(train_generator,
nb_epoch=nb_epoch,
samples_per_epoch=nb_train_samples,
validation_data=validation_generator,
nb_val_samples=nb_val_samples,
class_weight='auto')
# fine-tuning
setup_to_finetune(model)
history_ft = model.fit_generator(train_generator,
samples_per_epoch=nb_train_samples,
nb_epoch=nb_epoch,
validation_data=validation_generator,
nb_val_samples=nb_val_samples,
class_weight='auto')
model.save(args.output_model_file)
if args.plot:
plot_training(history_ft)
def define_model(IMAGE_DIMS, VEC_LEN, weight_dir):
i10 = Input(shape=IMAGE_DIMS)
i20 = Input(shape=IMAGE_DIMS)
i30 = Input(shape=IMAGE_DIMS)
t1 = Input(shape=(1, ))
t2 = Input(shape=(1, ))
i1 = Lambda(lambda x: tf.div(tf.subtract(tf.cast(x, tf.float32), 127.5),
127.5))(i10)
i2 = Lambda(lambda x: tf.div(tf.subtract(tf.cast(x, tf.float32), 127.5),
127.5))(i20)
i3 = Lambda(lambda x: tf.div(tf.subtract(tf.cast(x, tf.float32), 127.5),
127.5))(i30)
# i1 = tf.cast(i1,tf.float32)
# i2 = tf.cast(i2, tf.float32)
# i3 = tf.cast(i3, tf.float32)
#
# i1=tf.div(tf.subtract(i1,127.5),127.5)
# i2 = tf.div(tf.subtract(i2, 127.5), 127.5)
# i3 = tf.div(tf.subtract(i3, 127.5), 127.5)
print("[INFO] Weights restored from pre-trained InceptionV3!")
encoder = InceptionV3(weights=weight_dir, include_top=False)
pooling = GlobalAveragePooling2D()
def l2_normalize(x):
return K.expand_dims(K.l2_normalize(x, 1))
val = Lambda(l2_normalize, name='margin')()
BatchNormalization(axis=-1,
momentum=0.99,
epsilon=0.001,
center=True,
scale=True,
beta_initializer='zeros',
gamma_initializer='ones',
moving_mean_initializer='zeros',
moving_variance_initializer='ones',
beta_regularizer=None,
gamma_regularizer=None,
beta_constraint=None,
gamma_constraint=None)
output = Dense(VEC_LEN, activation='tanh', name='encoder_output')
o1 = encoder(i1)
o2 = encoder(i2)
o3 = encoder(i3)
# o1 = i1
# o2 = i2
# o3 = i3
o1 = pooling(o1) # 全局平均池化层
# o1 = BatchNormalization(o1)
o1 = output(o1) # 有1024个节点的全连接层
# o1 = NormLayer(o1)
#o1 = Dropout(0.1)(o1)
o2 = pooling(o2) # 全局平均池化层
# o2 = BatchNormalization(o2)
o2 = output(o2) # 有1024个节点的全连接层
#o2 = Dropout(0.1)(o2)
# o2 = NormLayer(o2)
o3 = pooling(o3) # 全局平均池化层
# o3 = BatchNormalization(o3)
o3 = output(o3) # 有1024个节点的全连接层
# o3 = NormLayer(o3)
#o3 = Dropout(0.1)(o3)
#print('[INFO] base_model_layers', len(encoder.layers))
def l2_normalize(x):
return K.expand_dims(K.l2_normalize(x, 1))
def l2_norm(x):
return K.sqrt(K.sum(K.square(x), 1))
def distance(inputs):
ap, an, margin, gthr = inputs
ap_l2n = K.sqrt(K.sum(K.square(ap), axis=1, keepdims=True))
an_l2n = K.sqrt(K.sum(K.square(an), axis=1, keepdims=True))
d = K.minimum((an_l2n - ap_l2n), margin)
# d=an_l2n
# g=ap_l2n
g = K.maximum(ap_l2n, gthr)
y = K.concatenate([d, g], axis=1)
return y
ap = Subtract()([o1, o2])
an = Subtract()([o1, o3])
val = Lambda(distance, name='margin')([ap, an, t1, t2])
# val = Concatenate()([d, g])
model = Model(inputs=[i10, i20, i30, t1, t2], outputs=val)
# K.clear_session()
return model
def initModel(name, input_shape):
model = Sequential()
if name == 'test':
model.add(Dense(512, input_shape=input_shape))
model.add(Activation("relu"))
model.add(Dense(7))
model.add(Activation("sigmoid"))
elif name == 'alexnet':
# 1st Convolutional Layer
model.add(
Conv2D(filters=96,
input_shape=input_shape,
kernel_size=(11, 11),
strides=(4, 4),
padding='valid'))
model.add(Activation('relu'))
# Max Pooling
model.add(
MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'))
# 2nd Convolutional Layer
model.add(
Conv2D(filters=256,
kernel_size=(11, 11),
strides=(1, 1),
padding='valid'))
model.add(Activation('relu'))
# Max Pooling
model.add(
MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'))
# 3rd Convolutional Layer
model.add(
Conv2D(filters=384,
kernel_size=(3, 3),
strides=(1, 1),
padding='valid'))
model.add(Activation('relu'))
# 4th Convolutional Layer
model.add(
Conv2D(filters=384,
kernel_size=(3, 3),
strides=(1, 1),
padding='valid'))
model.add(Activation('relu'))
# 5th Convolutional Layer
model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding='valid'))
model.add(Activation('relu'))
# Max Pooling
model.add(
MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'))
# Passing it to a Fully Connected layer
model.add(Flatten())
# 1st Fully Connected Layer
model.add(Dense(4096, input_shape=input_shape))
model.add(Activation('relu'))
# Add Dropout to prevent overfitting
model.add(Dropout(0.4))
# 2nd Fully Connected Layer
model.add(Dense(4096))
model.add(Activation('relu'))
# Add Dropout
model.add(Dropout(0.4))
# 3rd Fully Connected Layer
model.add(Dense(1000))
model.add(Activation('relu'))
# Add Dropout
model.add(Dropout(0.4))
# Output Layer
model.add(Dense(7))
model.add(Activation('softmax'))
elif name == 'VGG':
model.add(
Conv2D(64, (3, 3),
input_shape=input_shape,
padding='same',
activation='relu'))
model.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Flatten())
model.add(Dense(4096, input_shape=input_shape, activation='relu'))
model.add(Dense(4096, activation='relu'))
model.add(Dense(7, activation='softmax'))
elif name == 'Squeezenet':
img_input = Input(shape=input_shape)
x = Conv2D(64, (3, 3), strides=(2, 2), padding='valid',
name='conv1')(img_input)
x = Activation('relu', name='relu_conv1')(x)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool1')(x)
x = fire_module(x, fire_id=2, squeeze=16, expand=64)
x = fire_module(x, fire_id=3, squeeze=16, expand=64)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool3')(x)
x = fire_module(x, fire_id=4, squeeze=32, expand=128)
x = fire_module(x, fire_id=5, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool5')(x)
x = fire_module(x, fire_id=6, squeeze=48, expand=192)
x = fire_module(x, fire_id=7, squeeze=48, expand=192)
x = fire_module(x, fire_id=8, squeeze=64, expand=256)
x = fire_module(x, fire_id=9, squeeze=64, expand=256)
x = Dropout(0.5, name='drop9')(x)
x = Conv2D(7, (1, 1), padding='valid',
name='conv10')(x) #uses classes bias_initializer=class_bias
x = Activation('relu', name='relu_conv10')(x)
x = GlobalAveragePooling2D()(x)
x = Activation('softmax', name='loss')(x)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
inputs = img_input
model = Model(inputs, x, name='squeezenet')
elif name == 'deepsqueeze':
img_input = Input(shape=input_shape)
x = Conv2D(64, (3, 3), strides=(2, 2), padding='valid',
name='conv1')(img_input)
x = Activation('relu', name='relu_conv1')(x)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool1')(x)
x = fire_module(x, fire_id=2, squeeze=16, expand=64)
x = fire_module(x, fire_id=3, squeeze=16, expand=64)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool3')(x)
x = fire_module(x, fire_id=4, squeeze=32, expand=128)
x = fire_module(x, fire_id=5, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool5')(x)
x = fire_module(x, fire_id=6, squeeze=32, expand=128)
x = fire_module(x, fire_id=7, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool7')(x)
x = fire_module(x, fire_id=8, squeeze=32, expand=128)
x = fire_module(x, fire_id=9, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool9')(x)
x = fire_module(x, fire_id=14, squeeze=32, expand=128)
x = fire_module(x, fire_id=15, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool11')(x)
x = fire_module(x, fire_id=10, squeeze=48, expand=192)
x = fire_module(x, fire_id=11, squeeze=48, expand=192)
x = fire_module(x, fire_id=12, squeeze=64, expand=256)
x = fire_module(x, fire_id=13, squeeze=64, expand=256)
x = Dropout(0.5, name='drop13')(x)
x = Conv2D(7, (1, 1), padding='valid',
name='conv10')(x) # uses classes
x = Activation('relu', name='relu_conv10')(x)
# x = GlobalAveragePooling2D()(x)
x = GlobalAveragePooling2D(data_format='channels_last')(x)
x = Activation('softmax', name='loss')(x)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
inputs = img_input
model = Model(inputs, x, name='deepsqueeze')
elif name == 'son_of_deepsqueeze':
img_input = Input(shape=input_shape)
x = Conv2D(64, (3, 3), strides=(2, 2), padding='valid',
name='conv1')(img_input)
x = Activation('relu', name='relu_conv1')(x)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool1')(x)
x = fire_module(x, fire_id=2, squeeze=16, expand=64)
x = fire_module(x, fire_id=3, squeeze=16, expand=64)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool3')(x)
x = fire_module(x, fire_id=4, squeeze=32, expand=128)
x = fire_module(x, fire_id=5, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool5')(x)
x = fire_module(x, fire_id=6, squeeze=48, expand=192)
x = fire_module(x, fire_id=7, squeeze=48, expand=192)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool7')(x)
x = fire_module(x, fire_id=8, squeeze=64, expand=256)
x = fire_module(x, fire_id=9, squeeze=64, expand=256)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool9')(x)
x = fire_module(x, fire_id=14, squeeze=80, expand=320)
x = fire_module(x, fire_id=15, squeeze=80, expand=320)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool11')(x)
x = fire_module(x, fire_id=10, squeeze=96, expand=384)
x = fire_module(x, fire_id=11, squeeze=96, expand=384)
x = fire_module(x, fire_id=12, squeeze=112, expand=448)
x = fire_module(x, fire_id=13, squeeze=112, expand=448)
x = Dropout(0.5, name='drop13')(x)
x = Conv2D(7, (1, 1), padding='valid',
name='conv10')(x) # uses classes
x = Activation('relu', name='relu_conv10')(x)
# x = GlobalAveragePooling2D()(x)
x = GlobalAveragePooling2D(data_format='channels_last')(x)
x = Activation('softmax', name='loss')(x)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
inputs = img_input
model = Model(inputs, x, name='son_of_deepsqueeze')
elif name == 'the_big_squeeze': #snow_squeeze
img_input = Input(shape=input_shape)
x = Conv2D(64, (3, 3), strides=(2, 2), padding='valid',
name='conv1')(img_input)
x = Activation('relu', name='relu_conv1')(x)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool1')(x)
x = fire_module(x, fire_id=2, squeeze=16, expand=64)
x = fire_module(x, fire_id=3, squeeze=16, expand=64)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool3')(x)
x = fire_module(x, fire_id=4, squeeze=32, expand=128)
x = fire_module(x, fire_id=5, squeeze=32, expand=128)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool5')(x)
x = fire_module(x, fire_id=6, squeeze=48, expand=192)
x = fire_module(x, fire_id=7, squeeze=48, expand=192)
x = fire_module(x, fire_id=8, squeeze=64, expand=256)
x = fire_module(x, fire_id=9, squeeze=64, expand=256)
x = Dropout(0.5, name='drop9')(x)
x = Flatten()(x)
x = Dense(7, activation='softmax')(x)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
inputs = img_input
model = Model(inputs, x, name='the_big_squeeze')
elif name == 'inceptionV3':
InceptionV3(include_top=False, input_shape=input_shape, classes=7)
return model
from networks.UNet import UNet
from networks.VggNet import Vgg19Net, SmallVggNet
from training_loops.OptimizerHelper import VggOneBlockFunctional
model = Vgg19Net.build(IMAGE_DIMS[0], IMAGE_DIMS[1], IMAGE_DIMS[2], classes=3)
print('VGG19Net')
model.summary()
model = SmallVggNet.build(IMAGE_DIMS[0],
IMAGE_DIMS[1],
IMAGE_DIMS[2],
classes=3)
print('SmallVggNet')
model.summary()
model = InceptionV3(input_shape=IMAGE_DIMS, classes=3, weights=None)
print('InceptionV3')
model.summary()
model = UNet.build([IMAGE_DIMS[0], IMAGE_DIMS[1], 1], 3)
print('U-Net')
model.summary()
model = resnet50(IMAGE_DIMS, 3)
print('ResNet50')
model.summary()
model = VggOneBlockFunctional.build(IMAGE_DIMS[0],
IMAGE_DIMS[1],
IMAGE_DIMS[2],
classes=3)
def process_images_thru_tl(batch_size=32, input1=1024, input2=1024, model_name="vgg"):
with tf.device("/device:GPU:1"):
if model_name == "vgg":
model = VGG16(weights = "imagenet", include_top=False, input_shape = [224, 224, 3])
size = 224
elif model_name == "inceptionv3":
model = InceptionV3(weights = "imagenet", include_top=False, input_shape = [299, 299, 3])
size = 299
elif model_name == "resnet50":
model = ResNet50(weights = "imagenet", include_top=False, input_shape = [224, 224, 3])
size = 224
elif model_name == "mobilenet":
model = ResNet50(weights = "imagenet", include_top=False, input_shape = [224, 224, 3])
size = 224
elif model_name == "xception":
model = Xception(weights = "imagenet", include_top=False)
size = 299
print("%s %d %d %d" % (model_name, input1, input2, batch_size))
model.summary()
model.get_weights()
labels = []
batch = []
# input_ = Input(shape=(size,size,3),name = 'image_input')
output_ = model.output
with tf.device("/device:GPU:1"):
if model_name == "inceptionv3" or model_name == "xception":
x = GlobalAveragePooling2D(name='avg_pool')(output_)
else:
x = Flatten(name='flatten')(output_)
if input1 != 0:
x = Dense(input1, activation='relu', name='fc1')(x)
if input2 != 0:
x = Dense(input2, activation='relu', name='fc2')(x)
x = Dense(128, activation='softmax', name='predictions')(x)
for layer in model.layers:
layer.trainable = False
my_model = Model(inputs=output_, outputs=x)
my_model.summary()
if os.path.exists("weights_%s_%d_%d_%d.h5" % (model_name, input1, input2, batch_size)):
my_model.load_weights("weights_%s_%d_%d_%d.h5" % (model_name, input1, input2, batch_size))
my_model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
train_generator = get_train_data.train_generator(img_size=size, batch_size=batch_size)
valid_generator = get_train_data.valid_generator(img_size=size, batch_size=8)
csv_logger = CSVLogger('log.csv', append=True, separator=',')
my_model.fit_generator(
train_generator,
steps_per_epoch=2000,#1000
epochs=10,
validation_data=valid_generator,
validation_steps=200,#200
callbacks=[csv_logger])
my_model.save_weights("weights_%s_%d_%d_%d.h5" % (model_name, input1, input2, batch_size))
def fit(self,
learning_rate=1e-4,
epochs=5,
activation='relu',
dropout=0,
hidden_size=1024,
nb_layers=1,
include_class_weight=False,
batch_size=20,
save_model=False,
verbose=True,
fine_tuning=False,
NB_IV3_LAYERS_TO_FREEZE=279,
use_TPU=False,
transfer_model='Inception',
min_accuracy=None,
extract_SavedModel=False):
if transfer_model in ['Inception', 'Xception', 'Inception_Resnet']:
target_size = (299, 299)
else:
target_size = (224, 224)
#We expect the classes to be the name of the folders in the training set
self.categories = os.listdir(TRAIN_DIR)
"""
helper functions to to build tensors
inspired by https://www.tensorflow.org/tutorials/load_data/images
"""
def prepare_image(img_path):
#reshape the image
image = Image.open(img_path)
image = image.resize(target_size,
PIL.Image.BILINEAR).convert("RGB")
#convert the image into a numpy array, and expend to a size 4 tensor
image = img_to_array(image)
#rescale the pixels to a 0-1 range
image = image.astype(np.float32) / 255
return image
def generate_tuples(img_folder):
#loop through all the images
# Get all file names of images present in folder
classes = os.listdir(img_folder)
classes_paths = [
os.path.abspath(os.path.join(img_folder, i)) for i in classes
]
x = []
y = []
for i, j in enumerate(classes):
#for all the classes, get the list of pictures
img_paths = os.listdir(classes_paths[i])
img_paths = [
os.path.abspath(os.path.join(classes_paths[i], x))
for x in img_paths
]
for img_path in img_paths:
x.append(prepare_image(img_path))
y = y + [i]
return (np.array(x), np.array(y).astype(np.int32))
#get training data
(x_train,
y_train) = generate_tuples(parentdir + '/data/image_dataset/train')
(x_val, y_val) = generate_tuples(parentdir + '/data/image_dataset/val')
#train input_function: see https://colab.research.google.com/drive/1F8txK1JLXKtAkcvSRQz2o7NSTNoksuU2#scrollTo=abbwQQfH0td3
def get_training_dataset(batch_size=batch_size):
# Convert the inputs to a Dataset.
dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))
# Shuffle, repeat, and batch the examples.
dataset = dataset.shuffle(1000).repeat().batch(batch_size,
drop_remainder=True)
return dataset
def get_validation_dataset(batch_size=batch_size):
# Convert the inputs to a Dataset.
dataset = tf.data.Dataset.from_tensor_slices((x_val, y_val))
# Shuffle, repeat, and batch the examples.
dataset = dataset.shuffle(1000).repeat().batch(batch_size,
drop_remainder=True)
return dataset
#if we want stop training when no sufficient improvement in accuracy has been achieved
if min_accuracy is not None:
callback = EarlyStopping(monitor='acc', baseline=min_accuracy)
callback = [callback]
else:
callback = None
#load the pretrained model, without the classification (top) layers
if transfer_model == 'Xception':
base_model = Xception(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
elif transfer_model == 'Inception_Resnet':
base_model = InceptionResNetV2(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
elif transfer_model == 'Resnet':
base_model = ResNet50(weights='imagenet',
include_top=False,
input_shape=(224, 224, 3))
else:
base_model = InceptionV3(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
#Add the classification layers using Keras functional API
x = base_model.output
x = GlobalAveragePooling2D()(x)
for _ in range(nb_layers):
x = Dense(hidden_size, activation=activation)(
x) #Hidden layer for classification
if dropout > 0:
x = Dropout(rate=dropout)(x)
predictions = Dense(len(self.categories),
activation='softmax')(x) #Output layer
model = Model(inputs=base_model.input, outputs=predictions)
#Set only the top layers as trainable (if we want to do fine-tuning,
#we can train the base layers as a second step)
for layer in base_model.layers:
layer.trainable = False
#Define the optimizer and the loss, and compile the model
loss = 'sparse_categorical_crossentropy'
if use_TPU:
#if we want to try out the TPU, it looks like we currently need to use
#tensorflow optimizers...see https://stackoverflow.com/questions/52940552/valueerror-operation-utpu-140462710602256-varisinitializedop-has-been-marked
#...and https://www.youtube.com/watch?v=jgNwywYcH4w
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
model.compile(optimizer=optimizer,
loss=sparse_softmax_cross_entropy,
metrics=['acc'])
TPU_WORKER = 'grpc://' + os.environ['COLAB_TPU_ADDR']
model = tf.contrib.tpu.keras_to_tpu_model(
model,
strategy=tf.contrib.tpu.TPUDistributionStrategy(
tf.contrib.cluster_resolver.TPUClusterResolver(
TPU_WORKER)))
tf.logging.set_verbosity(tf.logging.INFO)
else:
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
model.compile(optimizer=optimizer, loss=loss, metrics=['acc'])
#if we want to weight the classes given the imbalanced number of images
if include_class_weight:
from sklearn.utils.class_weight import compute_class_weight
cls_train = self.categories
class_weight = compute_class_weight(class_weight='balanced',
classes=np.unique(cls_train),
y=cls_train)
else:
class_weight = None
steps_per_epoch = int(
sum([
len(files)
for r, d, files in os.walk(parentdir +
'/data/image_dataset/train')
]) / batch_size)
validation_steps = int(
sum([
len(files)
for r, d, files in os.walk(parentdir +
'/data/image_dataset/val')
]) / batch_size)
#Fit the model
if use_TPU:
history = model.fit(get_training_dataset,
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset,
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
else:
history = model.fit(get_training_dataset(),
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset(),
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
#Fine-tune the model, if we wish so
if fine_tuning and not model.stop_training:
print('============')
print('Begin fine-tuning')
print('============')
#declare the first layers as trainable
for layer in model.layers[:NB_IV3_LAYERS_TO_FREEZE]:
layer.trainable = False
for layer in model.layers[NB_IV3_LAYERS_TO_FREEZE:]:
layer.trainable = True
model.compile(optimizer=tf.train.AdamOptimizer(
learning_rate=learning_rate * 0.1),
loss=loss,
metrics=['acc'])
#Fit the model
if use_TPU:
history = model.fit(get_training_dataset,
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset,
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
else:
history = model.fit(get_training_dataset(),
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset(),
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
#Evaluate the model, just to be sure
self.fitness = history.history['val_categorical_accuracy'][-1]
#Save the model
if save_model:
if not os.path.exists(parentdir + '/data/trained_models'):
os.makedirs(parentdir + '/data/trained_models')
model.save(parentdir + '/data/trained_models/trained_model.h5')
print('Model saved!')
#save model in production format
if extract_SavedModel:
export_path = "./image_classifier/1/"
with K.get_session() as sess:
tf.saved_model.simple_save(
sess,
export_path,
inputs={'input_image': model.input},
outputs={t.name: t
for t in model.outputs})
else:
self.model = model
del history
del model
valid_datagen = ImageDataGenerator()
valid_batches = valid_datagen.flow_from_directory(DATATEST_PATH,
target_size=IMAGE_SIZE,
interpolation='bicubic',
class_mode='categorical',
shuffle=False,
batch_size=BATCH_SIZE)
# 輸出各類別的索引值
for cls, idx in train_batches.class_indices.items():
print('Class #{} = {}'.format(idx, cls))
# 以訓練好的 ResNet50 為基礎來建立模型，
# 捨棄 ResNet50 頂層的 fully connected layers
net = InceptionV3(include_top=False,
weights='imagenet',
input_tensor=None,
input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3))
x = net.output
x = Flatten()(x)
# 增加 DropOut layer
x = Dropout(0.5)(x)
# 增加 Dense layer，以 softmax 產生個類別的機率值
output_layer = Dense(NUM_CLASSES, activation='softmax', name='softmax')(x)
# 設定凍結與要進行訓練的網路層
net_final = Model(inputs=net.input, outputs=output_layer)
for layer in net_final.layers[:FREEZE_LAYERS]:
layer.trainable = False
for layer in net_final.layers[FREEZE_LAYERS:]:
img_width, img_height = 299, 299
epochs = 1
# Data Augmentation
train_datagen = ImageDataGenerator(rescale=1. / 255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
train_generator = train_datagen.flow_from_directory(
directory=train_data_dir,
target_size=[img_width, img_height],
class_mode='categorical')
# Step 2-1: Replace softmax Layer and add one dense layer
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
prediction = Dense(2, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=prediction)
for layer in model.layers:
layer.trainable = False
model.compile(loss='binary_crossentropy',
optimizer=optimizers.SGD(lr=1e-4, momentum=0.9),
metrics=['accuracy'])
model.fit_generator(train_generator,steps_per_epoch=10,epochs=epochs)
# Step 2-2: Unfreeze and train the top 2 layers
for layer in model.layers[:172]:
def predict_inceptionv3():
# InceptionV3 generators
test_batches = ImageDataGenerator(
preprocessing_function= \
applications.inception_v3.preprocess_input).flow_from_directory(
test2_path,
target_size=(299, 299),
batch_size=test_batch_size,
shuffle=False)
input_tensor = Input(shape=(299, 299, 3))
#Loading the model
model = InceptionV3(input_tensor=input_tensor,
weights='imagenet',
include_top=False)
# add a global spatial average pooling layer
x = model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(7, activation='softmax')(x)
# this is the model we will train
model = Model(inputs=model.input, outputs=predictions)
model.load_weights('modelInceptionV3.h5')
test_batches.reset()
test_labels = test_batches.classes
# Make predictions
predictions = model.predict_generator(test_batches,
steps=test_steps,
verbose=1)
# Declare a function for plotting the confusion matrix
def plot_confusion_matrix(cm,
classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
This function prints and plots the confusion matrix.
Normalization can be applied by setting `normalize=True`.
"""
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
print("Normalized confusion matrix")
else:
print('Confusion matrix, without normalization')
print(cm)
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
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.ylabel('True label')
plt.xlabel('Predicted label')
plt.tight_layout()
plt.savefig('confusion_matrix_InceptionV3.png')
cm = confusion_matrix(test_labels, predictions.argmax(axis=1))
cm_plot_labels = ['akiec', 'bcc', 'bkl', 'df', 'mel', 'nv', 'vasc']
plot_confusion_matrix(cm, cm_plot_labels)
recall = np.diag(cm) / np.sum(cm, axis=1)
precision = np.diag(cm) / np.sum(cm, axis=0)
print("Mean recall " + str(np.mean(recall).item()))
print("Mean precision " + str(np.mean(precision).item()))
print("Mean recall " + str(
recall_score(
test_labels, predictions.argmax(axis=1), average='weighted')))
print("Balanced Accuracy " +
str(balanced_accuracy_score(test_labels, predictions.argmax(
axis=1))))
print("Mean Precision " + str(
precision_score(
test_labels, predictions.argmax(axis=1), average='weighted')))
print("Mean f1 score " + str(
f1_score(test_labels, predictions.argmax(axis=1), average='weighted')))
file = open("InceptionV3_results.txt", "w+")
file.write("Mean recall " + str(np.mean(recall).item()) + "\n")
file.write("Mean precision " + str(np.mean(precision).item()) + "\n")
file.write("Mean recall " + str(
recall_score(
test_labels, predictions.argmax(axis=1), average='weighted')) +
"\n")
file.write(
"Balanced Accuracy " +
str(balanced_accuracy_score(test_labels, predictions.argmax(axis=1))) +
"\n")
file.write("Mean Precision " + str(
precision_score(
test_labels, predictions.argmax(axis=1), average='weighted')) +
"\n")
file.write("Mean f1 score " + str(
f1_score(test_labels, predictions.argmax(
axis=1), average='weighted')) + "\n")
file.close()
predicted_class_indices = np.argmax(predictions, axis=1)
labels = train_batches.class_indices
labels = dict((v, k) for k, v in labels.items())
predictions = [labels[k] for k in predicted_class_indices]
filenames = test_batches.filenames
results = pd.DataFrame({
"Filename": filenames,
"InceptionV3_Predictions": predictions
})
results.to_csv("test_prediction_inceptionv3.csv", index=False)
log_dir, checkpoints = keras_helper.ask_what_to_do_with_logfiles(
log_dir='./logs/inception_v3/',
ckpt="checkpoints/inception_v3/{epoch:02d}.ckpt")
dg = keras_helper.DataGenerators(
train_path="../../Dataset/sample_sizes/64_images/",
test_path="../../Dataset/test/",
enable_augmentation=True,
preprocessing_fn=preprocess_input)
train_generator, test_generator, tensorboard_generator = dg.create_data_generators(
299)
train_samples, test_samples = dg.get_samples()
batch_size = dg.get_batch_size()
model = keras_helper.create_model(InceptionV3(input_shape=(299, 299, 3),
include_top=False),
train_up_to=20)
model.compile(Adam(lr=0.0001),
loss="categorical_crossentropy",
metrics=["accuracy", keras_helper.top_3_acc])
print(model.summary())
tensorboard_callback = keras_callbacks.TensorBoardImage(log_dir=log_dir,
model=model,
DataGenerator=dg)
checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(checkpoints,
period=20)
model.fit_generator(train_generator,
steps_per_epoch=train_samples // batch_size,
def fit(self,
learning_rate=1e-4,
epochs=5,
activation='relu',
dropout=0,
hidden_size=1024,
nb_layers=1,
include_class_weight=False,
batch_size=20,
save_model=False,
verbose=True,
fine_tuning=False,
NB_IV3_LAYERS_TO_FREEZE=279,
use_TPU=False,
transfer_model='Inception',
min_accuracy=None,
extract_SavedModel=False):
#read the tfrecords data
TRAIN_DATA = tf.data.TFRecordDataset(['train.tfrecord'])
VAL_DATA = tf.data.TFRecordDataset(['val.tfrecord'])
print('Read the TFrecords')
if transfer_model in ['Inception', 'Xception', 'Inception_Resnet']:
target_size = (299, 299)
else:
target_size = (224, 224)
#We expect the classes to be the name of the folders in the training set
self.categories = os.listdir(TRAIN_DIR)
"""
helper functions to load tfrecords. Strongly inspired by
https://colab.research.google.com/github/GoogleCloudPlatform/training-data-analyst/blob/master/courses/fast-and-lean-data-science/07_Keras_Flowers_TPU_playground.ipynb#scrollTo=LtAVr-4CP1rp
"""
def read_tfrecord(example):
features = {
"image": tf.FixedLenFeature(
(), tf.string), # tf.string means byte string
"label": tf.FixedLenFeature((), tf.int64)
}
example = tf.parse_single_example(example, features)
image = tf.image.decode_jpeg(example['image'])
image = tf.cast(
image,
tf.float32) / 255.0 # convert image to floats in [0, 1] range
image = tf.image.resize_images(
image,
size=[*target_size],
method=tf.image.ResizeMethod.BILINEAR)
feature = tf.reshape(image, [*target_size, 3])
label = tf.cast(example['label'], tf.int32) # byte string
target = tf.one_hot(label, len(self.categories))
return feature, target
def get_training_dataset():
dataset = TRAIN_DATA.map(read_tfrecord)
dataset = dataset.cache()
dataset = dataset.repeat()
dataset = dataset.shuffle(1000)
dataset = dataset.batch(
batch_size,
drop_remainder=True) # drop_remainder needed on TPU
dataset = dataset.prefetch(
-1
) # prefetch next batch while training (-1: autotune prefetch buffer size)
return dataset
def get_validation_dataset():
dataset = VAL_DATA.map(read_tfrecord)
dataset = dataset.cache()
dataset = dataset.repeat()
dataset = dataset.shuffle(1000)
dataset = dataset.batch(
batch_size,
drop_remainder=True) # drop_remainder needed on TPU
dataset = dataset.prefetch(
-1
) # prefetch next batch while training (-1: autotune prefetch buffer size)
return dataset
#if we want stop training when no sufficient improvement in accuracy has been achieved
if min_accuracy is not None:
callback = EarlyStopping(monitor='categorical_accuracy',
baseline=min_accuracy)
callback = [callback]
else:
callback = None
#load the pretrained model, without the classification (top) layers
if transfer_model == 'Xception':
base_model = Xception(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
elif transfer_model == 'Inception_Resnet':
base_model = InceptionResNetV2(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
elif transfer_model == 'Resnet':
base_model = ResNet50(weights='imagenet',
include_top=False,
input_shape=(224, 224, 3))
else:
base_model = InceptionV3(weights='imagenet',
include_top=False,
input_shape=(299, 299, 3))
#Add the classification layers using Keras functional API
x = base_model.output
x = GlobalAveragePooling2D()(x)
for _ in range(nb_layers):
x = Dense(hidden_size, activation=activation)(
x) #Hidden layer for classification
if dropout > 0:
x = Dropout(rate=dropout)(x)
predictions = Dense(len(self.categories),
activation='softmax')(x) #Output layer
model = Model(inputs=base_model.input, outputs=predictions)
#Set only the top layers as trainable (if we want to do fine-tuning,
#we can train the base layers as a second step)
for layer in base_model.layers:
layer.trainable = False
#Define the optimizer and the loss, and compile the model
loss = 'categorical_crossentropy'
if use_TPU:
#if we want to try out the TPU, it looks like we currently need to use
#tensorflow optimizers...see https://stackoverflow.com/questions/52940552/valueerror-operation-utpu-140462710602256-varisinitializedop-has-been-marked
#...and https://www.youtube.com/watch?v=jgNwywYcH4w
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
tpu_optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
model.compile(optimizer=tpu_optimizer,
loss=loss,
metrics=['categorical_accuracy'])
TPU_WORKER = 'grpc://' + os.environ['COLAB_TPU_ADDR']
model = tf.contrib.tpu.keras_to_tpu_model(
model,
strategy=tf.contrib.tpu.TPUDistributionStrategy(
tf.contrib.cluster_resolver.TPUClusterResolver(
TPU_WORKER)))
tf.logging.set_verbosity(tf.logging.INFO)
else:
optimizer = Adam(lr=learning_rate)
model.compile(optimizer=optimizer,
loss=loss,
metrics=['categorical_accuracy'])
#if we want to weight the classes given the imbalanced number of images
if include_class_weight:
from sklearn.utils.class_weight import compute_class_weight
cls_train = self.categories
class_weight = compute_class_weight(class_weight='balanced',
classes=np.unique(cls_train),
y=cls_train)
else:
class_weight = None
steps_per_epoch = int(
sum([
len(files)
for r, d, files in os.walk(parentdir +
'/data/image_dataset/train')
]) / batch_size)
validation_steps = int(
sum([
len(files)
for r, d, files in os.walk(parentdir +
'/data/image_dataset/val')
]) / batch_size)
#Fit the model
if use_TPU:
history = model.fit(get_training_dataset,
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset,
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
else:
history = model.fit(get_training_dataset(),
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset(),
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
#Fine-tune the model, if we wish so
if fine_tuning and not model.stop_training:
print('============')
print('Begin fine-tuning')
print('============')
#declare the first layers as trainable
for layer in model.layers[:NB_IV3_LAYERS_TO_FREEZE]:
layer.trainable = False
for layer in model.layers[NB_IV3_LAYERS_TO_FREEZE:]:
layer.trainable = True
model.compile(optimizer=Adam(lr=learning_rate * 0.1),
loss=loss,
metrics=['categorical_accuracy'])
#Fit the model
if use_TPU:
history = model.fit(get_training_dataset,
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset,
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
else:
history = model.fit(get_training_dataset(),
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=get_validation_dataset(),
validation_steps=validation_steps,
verbose=verbose,
callbacks=callback,
class_weight=class_weight)
#Evaluate the model, just to be sure
self.fitness = history.history['val_categorical_accuracy'][-1]
#Save the model
if save_model:
if not os.path.exists(parentdir + '/data/trained_models'):
os.makedirs(parentdir + '/data/trained_models')
model.save(parentdir + '/data/trained_models/trained_model.h5')
print('Model saved!')
#save model in production format
if extract_SavedModel:
export_path = "./image_classifier/1/"
with K.get_session() as sess:
tf.saved_model.simple_save(
sess,
export_path,
inputs={'input_image': model.input},
outputs={t.name: t
for t in model.outputs})
else:
self.model = model
del history
del model
except:
model = Model(input=base_model.input,
output=base_model.get_layer('fc1').output)
image_size = (224, 224)
elif model_name == "resnet50":
base_model = ResNet50(weights=weights)
try:
model = Model(base_model.input,
base_model.get_layer('avg_pool').output)
except:
model = Model(input=base_model.input,
output=base_model.get_layer('avg_pool').output)
image_size = (224, 224)
elif model_name == "inceptionv3":
base_model = InceptionV3(include_top=include_top,
weights=weights,
input_tensor=Input(shape=(299, 299, 3)))
try:
model = Model(base_model.input,
base_model.get_layer('custom').output)
except:
model = Model(input=base_model.input,
output=base_model.get_layer('custom').output)
image_size = (299, 299)
elif model_name == "inceptionresnetv2":
base_model = InceptionResNetV2(include_top=include_top,
weights=weights,
input_tensor=Input(shape=(299, 299, 3)))
try:
model = Model(base_model.input,
base_model.get_layer('custom').output)
from tensorflow.python.keras import layers
from tensorflow.python.keras import Model
import wget
print("Inceptionv3 Weights is loading...")
wget.download("https://storage.googleapis.com/mledu-datasets/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5", "/tmp/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5")
print("Inceptionv3 Weights is loaded.")
from tensorflow.python.keras.applications.inception_v3 import InceptionV3
local_weights_file = '/tmp/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5'
pre_trained_model = InceptionV3(input_shape = (150, 150, 3),
include_top = False,
weights = None)
pre_trained_model.load_weights(local_weights_file)
for layer in pre_trained_model.layers:
layer.trainable = False
# pre_trained_model.summary()
last_layer = pre_trained_model.get_layer('mixed7')
print('last layer output shape: ', last_layer.output_shape)
last_output = last_layer.output
import tensorflow as tf
###############################################################################################################3
#Δήλωση
app = FastAPI()
#Δήλωση μοντέλων μηχανικής μάθησης και οι αντίστοιχες ρυθμίσεις τους
modelRes = ResNet50(weights='imagenet')
modelVGG = VGG16(weights='imagenet', include_top=True)
modelEfficient = EfficientNetB0(include_top=True, weights='imagenet')
modelEfficientB7 = EfficientNetB7(include_top=True, weights='imagenet')
modelNasNet = NASNetLarge(include_top=True, weights='imagenet')
modelMobileNet = MobileNetV2(weights="imagenet", include_top=True)
modelXception = Xception(include_top=True, weights="imagenet")
modelInception = InceptionV3(include_top=True,
weights="imagenet",
classifier_activation="softmax")
modelInRes = InceptionResNetV2(weights="imagenet", include_top=True)
# Settings
MIN_CONFIDENCE = 0.1 # Το ελάχιστο δυνατό confidence που δέχόμαστε απο τα μοντέλα.
# Τα URLs που θα απαντάει το κάθε μοντέλο
IMAGE_URL2 = "/v1/vision/resNet"
IMAGE_URL3 = "/v1/vision/vggNet"
IMAGE_URL4 = "/v1/vision/efficientNet"
IMAGE_URL5 = "/v1/vision/nasNet"
IMAGE_URL6 = "/v1/vision/mobileNet2"
IMAGE_URL7 = "/v1/vision/xceptionNet"
IMAGE_URL8 = "/v1/vision/efficientNetB7"
def run_model(args):
# Configure the memory optimizer
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
config = tf.ConfigProto()
config.graph_options.rewrite_options.memory_optimization = rewriter_config_pb2.RewriterConfig.SCHEDULING_HEURISTICS
#config.gpu_options.allow_growth=True
config.gpu_options.per_process_gpu_memory_fraction = 0.5
K.set_session(tf.Session(config=config))
num_classes = args.num_classes
batch_size = args.batch_size
model_name = args.model
if model_name == 'ResNet50':
model = ResNet50(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'ResNet101':
model = keras.applications.resnet.ResNet101(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'ResNet152':
model = ResNet152(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'VGG16':
model = VGG16(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'VGG19':
model = VGG19(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'Xception':
model = Xception(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'MobileNet':
model = MobileNet(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'MobileNetV2':
model = MobilenetV2(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
elif model_name == 'InceptionV3':
model = InceptionV3(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
else:
print('Running with ResNet50 -- the default model')
model = ResNet50(weights=None,
include_top=True,
input_shape=input_shape,
classes=num_classes)
execute_model(model, input_shape)
from tensorflow.python.keras import Input, layers
from tensorflow.python.keras import optimizers
from tensorflow.python.keras.models import Model
from tensorflow.python.keras.layers import add
from tensorflow.python.keras.preprocessing.sequence import pad_sequences
from tensorflow.python.keras.utils import to_categorical
import matplotlib.pyplot as plt
START = "startseq"
STOP = "endseq"
root_captioning = "D:\\datasets\\image_caption_data"
encode_model = InceptionV3(weights='imagenet')
encode_model = Model(encode_model.input, encode_model.layers[-2].output)
WIDTH = 299
HEIGHT = 299
OUTPUT_DIM = 2048
preprocess_input = tensorflow.keras.applications.inception_v3.preprocess_input
test_path = os.path.join(root_captioning, "data", f'Vocab.pkl')
with open(test_path, "rb") as fp:
vocab = pickle.load(fp)
idxtoword = {}
wordtoidx = {}
ix = 1
for w in vocab:
