def create_embedding():
base_cnn = resnet.ResNet50(weights="imagenet",
input_shape=target_shape + (3, ),
pooling="max",
include_top=False)
dense1 = layers.Dropout(0.3)(base_cnn.output)
output = layers.Dense(128)(dense1)
embedding = Model(base_cnn.input, output, name="Embedding")
trainable = False
for layer in base_cnn.layers:
if layer.name == "conv5_block3_1_conv":
trainable = True
layer.trainable = trainable
return embedding
def create_embedding():
base_cnn = resnet.ResNet50(weights="imagenet",
input_shape=target_shape + (3, ),
include_top=False)
flatten = layers.Flatten()(base_cnn.output)
dense1 = layers.Dense(512, activation="relu")(flatten)
dense1 = layers.BatchNormalization()(dense1)
dense1 = layers.Dropout(0.5)(dense1)
dense2 = layers.Dense(256, activation="relu")(dense1)
dense2 = layers.BatchNormalization()(dense2)
dense2 = layers.Dropout(0.5)(dense2)
output = layers.Dense(256)(dense2)
embedding = Model(base_cnn.input, output, name="Embedding")
trainable = False
for layer in base_cnn.layers:
if layer.name == "conv5_block1_out":
trainable = True
layer.trainable = trainable
return embedding
"""
## Setting up the embedding generator model
Our Siamese Network will generate embeddings for each of the images of the
triplet. To do this, we will use a ResNet50 model pretrained on ImageNet and
connect a few `Dense` layers to it so we can learn to separate these
embeddings.
We will freeze the weights of all the layers of the model up until the layer `conv5_block1_out`.
This is important to avoid affecting the weights that the model has already learned.
We are going to leave the bottom few layers trainable, so that we can fine-tune their weights
during training.
"""
base_cnn = resnet.ResNet50(
weights="imagenet", input_shape=target_shape + (3,), include_top=False
)
flatten = layers.Flatten()(base_cnn.output)
dense1 = layers.Dense(512, activation="relu")(flatten)
dense1 = layers.BatchNormalization()(dense1)
dense2 = layers.Dense(256, activation="relu")(dense1)
dense2 = layers.BatchNormalization()(dense2)
output = layers.Dense(256)(dense2)
embedding = Model(base_cnn.input, output, name="Embedding")
trainable = False
for layer in base_cnn.layers:
if layer.name == "conv5_block1_out":
trainable = True
def __call__(self, *args, **kwargs):
if self.model_name in ["VGG16", "vgg16"]:
pre_trained = vgg16.VGG16(include_top=False,
weights='imagenet',
input_shape=self.input_shape)
elif self.model_name in ["VGG19", "vgg19"]:
pre_trained = vgg19.VGG19(include_top=False,
weights='imagenet',
input_shape=self.input_shape)
elif self.model_name in ["MobileNet", "mobilenet"]:
pre_trained = mobilenet.MobileNet(include_top=False,
weights='imagenet',
input_shape=self.input_shape)
elif self.model_name in [
"MobileNetV2", "MobileNet_V2", "mobilenetv2", "mobilenet_v2",
"mobilenet_V2"
]:
pre_trained = mobilenet_v2.MobileNetV2(
include_top=False,
weights='imagenet',
input_shape=self.input_shape)
elif self.model_name in ["resnet50", "ResNet50"]:
pre_trained = resnet.ResNet50(include_top=False,
weights='imagenet',
input_shape=self.input_shape)
# elif self.model_name in ["EfficientNetB0", "efficientnetb0"]:
# pre_trained = efficientnet.EfficientNetB0(include_top=False, weights='imagenet', input_shape=self.input_shape)
#
# elif self.model_name in ["EfficientNetB5", "efficientnetb5"]:
# pre_trained = efficientnet.EfficientNetB5(include_top=False, weights='imagenet', input_shape=self.input_shape)
else:
print("Not exists {}".format(self.model_name))
return None
if self.extractor:
for layer in pre_trained.layers:
layer.trainable = False
if self.model_name in ["VGG16", "vgg16", "VGG19", "vgg19"]:
x = Flatten()(pre_trained.output)
x = Dense(1024, activation="relu",
kernel_initializer="he_normal")(x)
x = Dropout(0.5)(x)
x = Dense(1024, activation="relu",
kernel_initializer="he_normal")(x)
x = Dropout(0.5)(x)
else:
x = GlobalAveragePooling2D()(pre_trained.output)
x = Flatten()(x)
y = Dense(2, activation="softmax")(x)
model = Model(inputs=pre_trained.input, outputs=y)
model.layers[0]._name = "input"
return model
rotation_range=30,
zoom_range=[.8, 1.2],
rescale=1 / 255.)
gen = datagen.flow_from_directory(directory='./chars/',
target_size=(100, 100),
color_mode='grayscale',
batch_size=8,
interpolation='nearest')
a_file = open("model_loader.pkl", "rb")
dict_file = pickle.load(a_file)
a_file.close()
gen.class_indices = dict_file
model = resnet.ResNet50(input_shape=(100, 100, 1),
weights=None,
classes=gen.num_classes)
opt = Adam(learning_rate=0.0001)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.summary()
plot_model(model, show_shapes=True)
hist = model.fit_generator(gen, epochs=30)
model.save('model.h5', include_optimizer=False)
loss = hist.history['loss']
acc = hist.history['acc']
epochs = range(1, len(loss) + 1)
chosen_dist = tf.gather(dist, chosen_idx)
chosen_sim = tf.gather(sim, chosen_idx)
return (chosen_img, chosen_start, chosen_end, chosen_dist, chosen_sim)
# Code start
src = '/home/anhuynh/.keras/models/resnet50_weights_tf_dim_ordering_tf_kernels.h5_tf2'
dst = '/home/anhuynh/.keras/models/resnet50_weights_tf_dim_ordering_tf_kernels.h5'
shutil.copy(src,dst)
model = resnet.ResNet50()
model = tf.keras.Model(inputs=model.input, outputs=model.get_layer('avg_pool').output)
# model.summary()
num_stop = 1000
stop_flg = False
# load image
img_1_path = 'hit2.png'
gsv_dir = 'GSV_Step_1_3_7/'
gsv_names = os.listdir(gsv_dir)
gsv_names.sort()
# gsv_names = ['4.jpeg','5.jpeg']
y,
test_size=0.2,
random_state=42,
stratify=y)
# MinMax
print(np.min(X), np.max(X))
x_train /= -80
x_val /= -80
# 모델
input_tensor = Input(shape=X.shape[1:], dtype='float32', name='input')
model = resnet.ResNet50(input_tensor=input_tensor,
weights=None,
pooling='max',
classes=13)
model.summary()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
file_path = './project/mini/data/genre_model_resnet50_app.hdf5'
es = EarlyStopping(monitor='val_accuracy', patience=50)
cp = ModelCheckpoint(filepath=file_path,
monitor='val_accuracy',
save_best_only=True)
lr = ReduceLROnPlateau(monitor='val_accuracy', factor=0.5, patience=20)
history = model.fit(x_train,
y_train,