def efficient_model(input_shape=(224, 224, 3)):
model = EfficientNetB0(input_shape=input_shape,
weights=None) #can do B7 if want larger
input_layer = model.input
last_layer = model.layers[-2].output
x = Dense(1, activation='sigmoid')(last_layer)
return Model(inputs=input_layer, outputs=x)
def _build(self):
# inputs = keras.layers.Input((*self.input_dims, 1))
# x = keras.layers.Conv2D(filters=3, kernel_size=(1, 1), strides=(1, 1), name="initial_conv2d")(inputs)
# x = keras.layers.BatchNormalization(axis=3, epsilon=1.001e-5, name='initial_bn')(x)
# x = keras.layers.Activation('relu', name='initial_relu')(x)
#
engine = EfficientNetB0(include_top=False,
input_shape=(*self.input_dims[:2], 3),
classes=6)
x = keras.layers.GlobalAveragePooling2D(name='avg_pool')(engine.output)
x = keras.layers.Dropout(0.2)(x)
x = keras.layers.Dense(keras.backend.int_shape(x)[1],
activation="relu",
name="dense_hidden_1")(x)
x = keras.layers.Dropout(0.1)(x)
out = keras.layers.Dense(6, activation="sigmoid",
name='dense_output')(x)
self.model = keras.models.Model(inputs=engine.input, outputs=out)
self.model.compile(loss='binary_crossentropy',
optimizer='sgd',
metrics=['acc', 'mse'])
def _build(self):
self, model = EfficientNetB0(include_top=False,
input_shape=(*self.input_dims, 1),
classes=6)
self.model.compile(loss=weighted_log_loss,
optimizer=keras.optimizers.Adam(0.0),
metrics=[weighted_loss])
def build_efficient_net(n_classes, img_size, model_name):
input_layer = Input((img_size, img_size, 3))
eff_net = EfficientNetB0(weights=None,
include_top=False,
input_tensor=input_layer)
eff_net.trainable = True
head = vanilla_head(eff_net.output, n_classes)
return Model(input_layer, head, name=model_name)
def __init__(self,
model_type='vgg19',
layer='fc1',
weights='imagenet',
img_dims=(320, 240),
**kwargs):
"""Initializes the object feature detector module
efficientnet/probs
Args:
vgg19:fc2
"""
super().__init__(**kwargs)
self.weights = weights
self.layer = layer
self.session = tf.Session(graph=tf.Graph())
self.pre_input = None
self.img_dims = img_dims
self.queue = deque(maxlen=1)
with self.session.graph.as_default():
set_session(self.session)
if model_type == 'efficientnet':
from efficientnet.keras import EfficientNetB0
from efficientnet.keras import preprocess_input
self.pre_input = preprocess_input
base_model = EfficientNetB0(weights=self.weights)
self.xs, self.ys = 224, 224
if model_type == 'inveptionv3':
from keras.applications.inception_v3 import InceptionV3
from keras.applications.inception_v3 import preprocess_input
base_model = InceptionV3(weights=self.weights,
include_top=True)
self.pre_input = preprocess_input
self.xs, self.ys = 299, 299
if model_type == 'vgg19':
from keras.applications.vgg19 import VGG19
from keras.applications.vgg19 import preprocess_input
base_model = VGG19(weights=self.weights, include_top=True)
self.pre_input = preprocess_input
self.xs, self.ys = 224, 224
# allow for other output layers
print(base_model.summary())
self.model = Model(inputs=base_model.input,
outputs=base_model.get_layer(self.layer).output)
print('Feature extractor module setup complete.')
def _get_backbone_model(self):
backbone_model = None
if self.backbone_name == 'resnet18':
backbone_model = ResNet18(input_shape=self.input_shape, include_top=False, weights='imagenet')
if self.backbone_name == 'resnet34':
backbone_model = ResNet34(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'mobilenetv2':
backbone_model = MobileNetV2(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'efficientnetb0':
backbone_model = EfficientNetB0(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'efficientnetb1':
backbone_model = EfficientNetB1(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'efficientnetb3':
backbone_model = EfficientNetB3(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'efficientnetb5':
backbone_model = EfficientNetB5(input_shape=self.input_shape, include_top=False, weights='imagenet')
elif self.backbone_name == 'densenet':
backbone_model = DenseNet121(input_shape=self.input_shape, include_top=False, weights='imagenet')
return backbone_model
def __init__(self, model_size , imag_size, num_cls):
try:
if model_size == "B0": model = EfficientNetB0(weights = 'imagenet', input_shape = (imag_size,imag_size,3), include_top = False)
elif model_size == "B3": model = EfficientNetB3(weights = 'imagenet', input_shape = (imag_size,imag_size,3), include_top = False)
elif model_size == "B5": model = EfficientNetB5(weights = 'imagenet', input_shape = (imag_size,imag_size,3), include_top = False)
elif model_size == "B7": model = EfficientNetB7(weights = 'imagenet', input_shape = (imag_size,imag_size,3), include_top = False)
ENet_out = model.output
ENet_out = Flatten()(ENet_out)
Hidden1_in = Dense(1024, activation="relu")(ENet_out)
Hidden1_in = Dropout(0.5)(Hidden1_in)
predictions = Dense(units = num_cls, activation="softmax")(Hidden1_in)
self.model_f = Model(input = model.input, output = predictions)
self.model_f.compile(optimizers.Adam(lr=0.0001, beta_1=0.9, beta_2=0.999, epsilon=1e-08), loss='categorical_crossentropy', metrics=[metrics.mae, metrics.categorical_accuracy])
except:
print("only B0/B3/B5/B7 allowed")
if not os.path.isdir(result_path):
os.mkdir(DEFAULT_IMAGE_DIR + '/result')
os.mkdir(DEFAULT_IMAGE_DIR + '/cropped')
################# LOAD MODELS #######################
#### MASK-R-CNN
mrcnn_model = modellib.MaskRCNN(mode="inference",
config=config,
model_dir=DEFAULT_LOGS_DIR)
weights_path = DEFAULT_MRCNN_MODEL_DIR
mrcnn_model.load_weights(weights_path, by_name=True)
##### EFFICIENTNET
efficient_net = EfficientNetB0(weights='imagenet',
input_shape=(32, 32, 3),
include_top=False,
pooling='max')
eff_model = Sequential()
eff_model.add(efficient_net)
eff_model.add(Dense(units=120, activation='relu'))
eff_model.add(Dense(units=120, activation='relu'))
eff_model.add(Dense(units=1, activation='sigmoid'))
eff_model.summary()
eff_model.compile(optimizer=Adam(lr=0.0001),
loss='binary_crossentropy',
metrics=['accuracy'])
eff_model.load_weights(DEFAULT_EFF_MODEL_DIR)
#############################################################################3
#name: Image Classification
#description: Image classification based of Efficientnet model
#language: python
#input: file file
#output: map classes [Detected classes with probabilities]
#tags: demo, panel, files, efficientnet
#condition: file.isFile && file.size < 1e6 && (file.name.endsWith("jpg") || file.name.endsWith("jpeg") || file.name.endsWith("png"))
#help-url: https://github.com/qubvel/efficientnet
import numpy as np
from skimage.io import imread
from efficientnet.keras import EfficientNetB0
from keras.applications.imagenet_utils import decode_predictions
from efficientnet.keras import center_crop_and_resize, preprocess_input
image = imread(file)
model = EfficientNetB0(weights='imagenet')
image_size = model.input_shape[1]
_image = center_crop_and_resize(image, image_size=image_size)
_image = preprocess_input(_image)
_image = np.expand_dims(_image, 0)
predicted = model.predict(_image)
predicted = decode_predictions(predicted)[0]
classes = {}
for p in predicted:
classes[p[1]] = float(p[2])
verbose=1,
save_best_only=True,
mode="max",
)
lr_reduce = ReduceLROnPlateau(
monitor="val_loss",
factor=np.sqrt(0.1),
patience=5,
verbose=1,
cooldown=0,
min_lr=0.5e-6,
)
callbacks = [checkpoint, lr_reduce]
conv_m = EfficientNetB0(weights="imagenet",
include_top=False,
input_shape=(size, size, 3))
conv_m.trainable = False
model = Sequential()
model.add(conv_m)
model.add(AveragePooling2D(pool_size=(7, 7)))
model.add(Flatten())
model.add(Dense(32, activation="relu"))
model.add(BatchNormalization())
model.add(Dropout(0.5))
model.add(Dense(5, activation="softmax"))
model.compile(
loss="categorical_crossentropy",
optimizer=SGD(lr=0.1, momentum=0.9),
metrics=["accuracy"],
target_size=(img_rows, img_cols),
batch_size=batch_size,
class_mode='categorical')#多分类; 'binary')
print("train_generator.filenames",train_generator.filenames)# 按顺序输出文件的名字
print("train_generator.class_indices", train_generator.class_indices) #输出对应的标签文件夹
validation_generator = test_datagen.flow_from_directory(
test_data_dir,
target_size=(img_rows, img_cols),
batch_size=batch_size,
class_mode='categorical')#多分类; 'binary')
print("validation_generator.filenames",validation_generator.filenames)# 按顺序输出文件的名字
print("validation_generator.class_indices", validation_generator.class_indices) #输出对应的标签文件夹
##################训练网络模型###########################
# build model
base_model = EfficientNetB0(input_shape=(img_rows, img_cols, 3), weights='imagenet', include_top=False)
x = keras.layers.GlobalAveragePooling2D()(base_model.output)
output = keras.layers.Dense(num_classes, activation='softmax')(x)
model = keras.models.Model(inputs=[base_model.input], outputs=[output])
# initiate optimizer
opt = keras.optimizers.SGD(lr=0.001) # keras.optimizers.Adam(lr=0.001)
model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy']) # optimizer='SGD'
# train
history = model.fit_generator(
train_generator,
steps_per_epoch=nb_train_samples // batch_size, # typically steps_per_epoch= dataset samples 3120 / batch_size 4
epochs=epochs, # finacal epoches
validation_data=validation_generator,
validation_steps=nb_validation_samples// batch_size, # typically validation_steps = validation dataset samples 780 / batch_size 4
verbose = 2,
)
json_file.close()
model = model_from_json(model_json)
model.load_weights(args["weights"])
model.compile(
optimizer=Adam(lr=0.0001),
loss="binary_crossentropy",
metrics=["accuracy"],
)
break
elif choice in ("n", "N"):
sys.exit(0)
else:
# initializeaza si compileaza modelul EfficientNet (foloseste transfer learning)
efficient_net = EfficientNetB0(weights="imagenet",
input_shape=(256, 128, 3),
include_top=False,
pooling="max")
model = Sequential()
model.add(efficient_net)
model.add(Dense(units=120, activation="relu"))
model.add(Dense(units=120, activation="relu"))
model.add(Dense(units=2, activation="softmax"))
model.compile(optimizer=Adam(lr=0.0001),
loss="binary_crossentropy",
metrics=["accuracy"])
if not os.path.isfile(args["model"]):
model_json = model.to_json()
with open(args["model"], "w") as json_file:
json_file.write(model_json)
model.summary()
x_test, y_test = load_data(dataset_test_dst)
print("x_train.shape = {}".format(x_train.shape))
print("x_test.shape = {}".format(x_test.shape))
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255
# one-hot encoding
num_cls = 5
Y_train = np_utils.to_categorical(y_train, num_cls)
Y_test = np_utils.to_categorical(y_test, num_cls)
#------------------------------------end of importing data-----------------------------#
#------------------------------------start building model-----------------------------#
model = EfficientNetB0(weights='imagenet',
input_shape=(imag_size, imag_size, 3),
include_top=False)
ENet_out = model.output
ENet_out = Flatten()(ENet_out)
Hidden1_in = Dense(1024, activation="relu")(ENet_out)
Hidden1_in = Dropout(0.5)(Hidden1_in)
predictions = Dense(units=num_cls, activation="softmax")(Hidden1_in)
model_f = Model(input=model.input, output=predictions)
model_f.compile(optimizers.Adam(lr=0.0001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08),
loss='categorical_crossentropy',
def setup():
return EfficientNetB0(weights='imagenet')
model_split = 'block2a_expand_conv'
layer_idx = [
i for i in range(len(model.layers))
if model.layers[i].name == model_split
][0]
for i, layer in enumerate(model.layers):
if i < layer_idx or isinstance(layer, layers.BatchNormalization):
layer.trainable = False
else:
layer.trainable = True
else:
base_model = EfficientNetB0(include_top=False,
weights='imagenet',
input_shape=(None, None, 3),
pooling='avg')
x = base_model.output
predictions = layers.Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=predictions)
if freeze:
for layer in base_model.layers:
layer.trainable = False
else:
model_split = 'block2a_expand_conv'
layer_idx = [
i for i in range(len(model.layers))
opt = Adam(lr=INIT_LR)
reduce_lr = ReduceLROnPlateau(monitor='val_loss',
factor=0.1,
patience=3,
min_lr=1e-10,
verbose=1,
cooldown=2)
model_checkpoint = ModelCheckpoint('EFNB3Weight.h5',
monitor='val_loss',
save_best_only=True,
period=3)
input_tensor = Input(shape=(28, 28, 3))
model = EfficientNetB0(weights=None,
include_top=False,
input_tensor=input_tensor)
x = model.output
x = Flatten()(x)
x = Dense(1024, activation="relu")(x)
x = Dropout(0.5)(x)
predictions = Dense(units=10, activation="softmax")(x)
model_f = Model(input=model.input, output=predictions)
model_f.compile(Adam(lr=INIT_LR),
loss='categorical_crossentropy',
metrics=['accuracy'])
print(model.summary())
# train the network
