filters=16,
kernel_size=[3, 3],
strides=[2, 2],
padding="same",
dilation_rate=[1, 1],
kernel_initializer=Constant(
np.load(
'weights_1000/siamese_neural_congas_1_Mixed_6a_Branch_1_Conv2d_1c_3x3_Conv2D_weights'
).transpose(1, 2, 3, 0)),
bias_initializer=Constant(
np.load(
'weights_1000/siamese_neural_congas_1_Mixed_6a_Branch_1_Conv2d_1c_3x3_Conv2D_bias'
)))(relu2_4)
relu2_5 = ReLU(max_value=6.)(conv2_5)
maxpool2_1 = MaxPool2D(pool_size=[3, 3], strides=[2, 2],
padding='same')(relu1_3)
concat2_1 = Concatenate(axis=3)([relu2_2, relu2_5, maxpool2_1])
# Block_03
conv3_1 = Conv2D(
filters=32,
kernel_size=[1, 1],
strides=[1, 1],
padding="same",
dilation_rate=[1, 1],
kernel_initializer=Constant(
np.load(
'weights_1000/siamese_neural_congas_1_Mixed7a_Branch_0_Conv2d_0a_1x1_Conv2D_weights'
).transpose(1, 2, 3, 0)),
bias_initializer=Constant(
mnist = tf.keras.datasets.mnist
# Разделяем данные на обучающие и тестовые
(feature_train, label_train), (feature_test, label_test) = mnist.load_data()
# Нормировка данных изображений
feature_train, feature_test = feature_train / 255, feature_test / 255
feature_train, feature_test = np.expand_dims(
feature_train, axis=-1), np.expand_dims(feature_test, axis=-1)
# Инициализируем модель
model = Sequential()
# Создаем слои
model.add(Input(shape=(28, 28, 1)))
model.add(Conv2D(32, (3, 3), padding="valid", activation=tf.nn.relu))
model.add(MaxPool2D((2, 2), (2, 2)))
model.add(Dropout(0.5))
model.add(Conv2D(64, (3, 3), padding="valid", activation=tf.nn.relu))
model.add(MaxPool2D((2, 2), (2, 2)))
model.add(Dropout(0.5))
model.add(Conv2D(128, (3, 3), padding="valid", activation=tf.nn.relu))
model.add(Flatten())
model.add(Dense(10, activation=tf.nn.softmax))
# Компилируем модель
model.compile(optimizer="adam",
loss="sparse_categorical_crossentropy",
metrics=["accuracy"])
# Обучаем модель
model.fit(feature_train, label_train, epochs=5)
def TBPP_VGG16(
config,
num_predictions=10,
is_training=True,
):
model_config = config["model"]
if is_training:
input_shape = (None, None, 3)
else:
input_shape = (model_config["input_size"], model_config["input_size"],
3)
num_classes = 2 # 1 for text and 1 for background
l2_reg = model_config["l2_regularization"]
kernel_initializer = model_config["kernel_initializer"]
default_boxes_config = model_config["default_boxes"]
extra_box_for_ar_1 = model_config["extra_box_for_ar_1"]
input_tensor = Input(shape=input_shape)
input_tensor = ZeroPadding2D(padding=(2, 2))(input_tensor)
# construct the base network and extra feature layers
base_network = VGG16(input_tensor=input_tensor,
classes=num_classes,
weights='imagenet',
include_top=False)
base_network = Model(inputs=base_network.input,
outputs=base_network.get_layer('block5_conv3').output)
base_network.get_layer("input_1")._name = "input"
for layer in base_network.layers:
if "pool" in layer.name:
new_name = layer.name.replace("block", "")
new_name = new_name.split("_")
new_name = f"{new_name[1]}{new_name[0]}"
else:
new_name = layer.name.replace("conv", "")
new_name = new_name.replace("block", "conv")
base_network.get_layer(layer.name)._name = new_name
base_network.get_layer(layer.name)._kernel_initializer = "he_normal"
base_network.get_layer(layer.name)._kernel_regularizer = l2(l2_reg)
layer.trainable = False # each layer of the base network should not be trainable
def conv_block_1(x,
filters,
name,
padding='valid',
dilation_rate=(1, 1),
strides=(1, 1)):
return Conv2D(filters,
kernel_size=(3, 3),
strides=strides,
activation='relu',
padding=padding,
dilation_rate=dilation_rate,
kernel_initializer=kernel_initializer,
kernel_regularizer=l2(l2_reg),
name=name)(x)
def conv_block_2(x,
filters,
name,
padding='valid',
dilation_rate=(1, 1),
strides=(1, 1)):
return Conv2D(filters,
kernel_size=(1, 1),
strides=strides,
activation='relu',
padding=padding,
dilation_rate=dilation_rate,
kernel_initializer=kernel_initializer,
kernel_regularizer=l2(l2_reg),
name=name)(x)
pool5 = MaxPool2D(pool_size=(3, 3),
strides=(1, 1),
padding="same",
name="pool5")(base_network.get_layer('conv5_3').output)
fc6 = conv_block_1(x=pool5,
filters=1024,
padding="same",
dilation_rate=(6, 6),
name="fc6")
fc7 = conv_block_2(x=fc6, filters=1024, padding="same", name="fc7")
conv8_1 = conv_block_2(x=fc7, filters=256, padding="valid", name="conv8_1")
conv8_2 = conv_block_1(x=conv8_1,
filters=512,
padding="same",
strides=(2, 2),
name="conv8_2")
conv9_1 = conv_block_2(x=conv8_2,
filters=128,
padding="valid",
name="conv9_1")
conv9_2 = conv_block_1(x=conv9_1,
filters=256,
padding="same",
strides=(2, 2),
name="conv9_2")
conv10_1 = conv_block_2(x=conv9_2,
filters=128,
padding="valid",
name="conv10_1")
conv10_2 = conv_block_1(x=conv10_1,
filters=256,
padding="valid",
name="conv10_2")
conv11_1 = conv_block_2(x=conv10_2,
filters=128,
padding="valid",
name="conv11_1")
conv11_2 = conv_block_1(x=conv11_1,
filters=256,
padding="valid",
name="conv11_2")
model = Model(inputs=base_network.input, outputs=conv11_2)
# construct the prediction layers (conf, loc, & default_boxes)
scales = np.linspace(default_boxes_config["min_scale"],
default_boxes_config["max_scale"],
len(default_boxes_config["layers"]))
mbox_conf_layers = []
mbox_loc_layers = []
mbox_quad_layers = []
for i, layer in enumerate(default_boxes_config["layers"]):
num_default_boxes = get_number_default_boxes(
layer["aspect_ratios"], extra_box_for_ar_1=extra_box_for_ar_1)
x = model.get_layer(layer["name"]).output
layer_name = layer["name"]
# conv4_3 has different scales compared to other feature map layers
if layer_name == "conv4_3":
layer_name = f"{layer_name}_norm"
x = L2Normalization(gamma_init=20, name=layer_name)(x)
layer_mbox_conf = Conv2D(filters=num_default_boxes * num_classes,
kernel_size=(3, 5),
padding='same',
kernel_initializer=kernel_initializer,
kernel_regularizer=l2(l2_reg),
name=f"{layer_name}_mbox_conf")(x)
layer_mbox_conf_reshape = Reshape(
(-1, num_classes),
name=f"{layer_name}_mbox_conf_reshape")(layer_mbox_conf)
layer_mbox_loc = Conv2D(filters=num_default_boxes * 4,
kernel_size=(3, 5),
padding='same',
kernel_initializer=kernel_initializer,
kernel_regularizer=l2(l2_reg),
name=f"{layer_name}_mbox_loc")(x)
layer_mbox_loc_reshape = Reshape(
(-1, 4), name=f"{layer_name}_mbox_loc_reshape")(layer_mbox_loc)
layer_mbox_quad = Conv2D(filters=num_default_boxes * 8,
kernel_size=(3, 5),
padding='same',
kernel_initializer=kernel_initializer,
kernel_regularizer=l2(l2_reg),
name=f"{layer_name}_mbox_quad")(x)
layer_mbox_quad_reshape = Reshape(
(-1, 8), name=f"{layer_name}_mbox_quad_reshape")(layer_mbox_quad)
mbox_conf_layers.append(layer_mbox_conf_reshape)
mbox_loc_layers.append(layer_mbox_loc_reshape)
mbox_quad_layers.append(layer_mbox_quad_reshape)
# concentenate class confidence predictions from different feature map layers
mbox_conf = Concatenate(axis=-2, name="mbox_conf")(mbox_conf_layers)
mbox_conf_softmax = Activation('softmax',
name='mbox_conf_softmax')(mbox_conf)
# concentenate object location predictions from different feature map layers
mbox_loc = Concatenate(axis=-2, name="mbox_loc")(mbox_loc_layers)
# concentenate object quad predictions from different feature map layers
mbox_quad = Concatenate(axis=-2, name="mbox_quad")(mbox_quad_layers)
if is_training:
# concatenate confidence score predictions, bounding box predictions, and default boxes
predictions = Concatenate(axis=-1, name='predictions')(
[mbox_conf_softmax, mbox_loc, mbox_quad])
return Model(inputs=base_network.input, outputs=predictions)
mbox_default_boxes_layers = []
for i, layer in enumerate(default_boxes_config["layers"]):
num_default_boxes = get_number_default_boxes(
layer["aspect_ratios"], extra_box_for_ar_1=extra_box_for_ar_1)
x = model.get_layer(layer["name"]).output
layer_name = layer["name"]
layer_default_boxes = DefaultBoxes(
image_shape=input_shape,
scale=scales[i],
next_scale=scales[i + 1]
if i + 1 <= len(default_boxes_config["layers"]) - 1 else 1,
aspect_ratios=layer["aspect_ratios"],
variances=default_boxes_config["variances"],
extra_box_for_ar_1=extra_box_for_ar_1,
name=f"{layer_name}_default_boxes")(x)
layer_default_boxes_reshape = Reshape(
(-1, 8),
name=f"{layer_name}_default_boxes_reshape")(layer_default_boxes)
mbox_default_boxes_layers.append(layer_default_boxes_reshape)
# concentenate default boxes from different feature map layers
mbox_default_boxes = Concatenate(
axis=-2, name="mbox_default_boxes")(mbox_default_boxes_layers)
predictions = Concatenate(axis=-1, name='predictions')(
[mbox_conf_softmax, mbox_loc, mbox_quad, mbox_default_boxes])
decoded_predictions = DecodeTBPPPredictions(
input_size=model_config["input_size"],
num_predictions=num_predictions,
name="decoded_predictions")(predictions)
return Model(inputs=base_network.input, outputs=decoded_predictions)
def create_model_CNN(data, labels, class_names, test_size, seed, model_path):
"""Создать и обучить модель CNN, сохранить по пути model_path
Параметры:
data (list): список даных для тренировки НС
labels (list): список меток соответствующих данным
class_names (list): названия классов
test_size (float): процент данных для тестирования [0, 1]
seed (int): параметр для ГСЧ
model_path (list): путь, куда сохранять модель
"""
# Разделение выборки на тренировочную и тестовую
(train_data, test_data, train_labels,
test_labels) = train_test_split(np.array(data),
np.array(labels),
test_size=test_size,
random_state=seed)
# Определение архитектуры сверточной НС
# с двумя сверточными (Conv) и двумя полносвязными слоями (Dense)
model = Sequential()
# Сверточный слой с 32 фильтрами и ядром 3x3, ф-ия активации relu, размерность входных данных (32, 32, 3)
model.add(
Conv2D(filters=32,
kernel_size=(3, 3),
activation='relu',
input_shape=(32, 32, 3)))
# Слой субдискретизации для снижения размерности в 2 раза по каждой оси
model.add(MaxPool2D(pool_size=(2, 2)))
# Слой исключения нейронов для предотвращения переобучения
model.add(Dropout(0.3))
# Второй аналогичный сверточный слой
model.add(Conv2D(filters=64, kernel_size=(3, 3), activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Dropout(0.4))
# Слой, переводящий многомерные данные в одномерный массив
model.add(Flatten())
# Два скрытых полносвязных слоя с Dropout и выходной слой
model.add(Dense(128, activation='sigmoid'))
model.add(Dropout(0.25))
model.add(Dense(64, activation='sigmoid'))
model.add(Dropout(0.2))
model.add(Dense(3, activation='softmax'))
# Компилируем модель, используя Adam как оптимизатор и категориальную
# кросс-энтропию в качестве функции потерь
print("[INFO] Компиляция модели...")
optimizer = Adam(lr=0.001, beta_1=0.9, beta_2=0.999)
model.compile(loss="categorical_crossentropy",
optimizer=optimizer,
metrics=["accuracy"])
model.summary()
# Обучаем модель, epochs - количество эпох, batch_size - контролирует размер пакета данных для передачи по сети
print("[INFO] Обучение модели...")
history = model.fit(train_data,
train_labels,
validation_data=(test_data, test_labels),
epochs=15,
batch_size=32)
print("[INFO] Оценка модели...")
estimate_model(model, test_data, test_labels, class_names, history)
print("[INFO] Сериализация модели...")
model.save(model_path)
def create_model_cnn(params):
model = Sequential()
print("Training with params {}".format(params))
# (batch_size, timesteps, data_dim)
# x_train, y_train = get_data_cnn(df, df.head(1).iloc[0]["timestamp"])[0:2]
conv2d_layer1 = Conv2D(
params["conv2d_layers"]["conv2d_filters_1"],
params["conv2d_layers"]["conv2d_kernel_size_1"],
strides=params["conv2d_layers"]["conv2d_strides_1"],
kernel_regularizer=regularizers.l2(
params["conv2d_layers"]["kernel_regularizer_1"]),
padding='valid',
activation="relu",
use_bias=True,
kernel_initializer='glorot_uniform',
input_shape=(params['input_dim_1'], params['input_dim_2'],
params['input_dim_3']))
model.add(conv2d_layer1)
if params["conv2d_layers"]['conv2d_mp_1'] == 1:
model.add(MaxPool2D(pool_size=2))
model.add(Dropout(params['conv2d_layers']['conv2d_do_1']))
if params["conv2d_layers"]['layers'] == 'two':
conv2d_layer2 = Conv2D(
params["conv2d_layers"]["conv2d_filters_2"],
params["conv2d_layers"]["conv2d_kernel_size_2"],
strides=params["conv2d_layers"]["conv2d_strides_2"],
kernel_regularizer=regularizers.l2(
params["conv2d_layers"]["kernel_regularizer_2"]),
padding='valid',
activation="relu",
use_bias=True,
kernel_initializer='glorot_uniform')
model.add(conv2d_layer2)
if params["conv2d_layers"]['conv2d_mp_2'] == 1:
model.add(MaxPool2D(pool_size=2))
model.add(Dropout(params['conv2d_layers']['conv2d_do_2']))
model.add(Flatten())
model.add(Dense(params['dense_layers']["dense_nodes_1"],
activation='relu'))
model.add(Dropout(params['dense_layers']['dense_do_1']))
if params['dense_layers']["layers"] == 'two':
model.add(
Dense(params['dense_layers']["dense_nodes_2"],
activation='relu',
kernel_regularizer=params['dense_layers']
["kernel_regularizer_1"]))
model.add(Dropout(params['dense_layers']['dense_do_2']))
model.add(Dense(3, activation='softmax'))
if params["optimizer"] == 'rmsprop':
optimizer = optimizers.RMSprop(lr=params["lr"])
elif params["optimizer"] == 'sgd':
optimizer = optimizers.SGD(lr=params["lr"],
decay=1e-6,
momentum=0.9,
nesterov=True)
elif params["optimizer"] == 'adam':
optimizer = optimizers.Adam(learning_rate=params["lr"],
beta_1=0.9,
beta_2=0.999,
amsgrad=False)
model.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy', f1_metric])
# from keras.utils.vis_utils import plot_model use this too for diagram with plot
model.summary(print_fn=lambda x: print(x + '\n'))
return model
dtype='int32')
bm_input = Input(
shape=(MAX_SEQUENCE_LENGTH, ),
dtype='int32')
embedding_layer = Embedding(
MAX_NUM_WORDS, NUM_EMBEDDING_DIM)
top_embedded = embedding_layer(
top_input)
bm_embedded = embedding_layer(
bm_input)
reshape = Reshape((MAX_SEQUENCE_LENGTH, NUM_EMBEDDING_DIM, 1))(top_embedded)
reshape_1 = Reshape((MAX_SEQUENCE_LENGTH, NUM_EMBEDDING_DIM, 1))(bm_embedded)
conv_0 = Conv2D(num_filters, kernel_size=(filter_sizes[0], NUM_EMBEDDING_DIM), padding='valid', kernel_initializer='normal', activation='relu')(reshape)
conv_1 = Conv2D(num_filters, kernel_size=(filter_sizes[1], NUM_EMBEDDING_DIM), padding='valid', kernel_initializer='normal', activation='relu')(reshape_1)
maxpool_0 = MaxPool2D(pool_size=(MAX_SEQUENCE_LENGTH - filter_sizes[0] + 1, 1), strides=(1,1), padding='valid')(conv_0)
maxpool_1 = MaxPool2D(pool_size=(MAX_SEQUENCE_LENGTH - filter_sizes[1] + 1, 1), strides=(1,1), padding='valid')(conv_1)
concatenated_tensor = Concatenate(axis=1)([maxpool_0, maxpool_1])
flatten = Flatten()(concatenated_tensor)
dropout = Dropout(drop)(flatten)
top_input_bt = Input(
shape=(768, ),
dtype='float32')
bm_input_bt = Input(
shape=(768, ),
dtype='float32')
top_embedded_bt = Reshape((1, 768, ))(top_input_bt)
bm_embedded_bt = Reshape((1, 768, ))(bm_input_bt)
def keras_model_build(input_size=(224, 224, 3)): #480 480 3
# 输入
input = Input(shape=input_size, name='input')
x = Conv2D(input_shape=input_size,
filters=64,
kernel_size=(7, 7),
activation='relu',
padding='same',
strides=(2, 2))(input)
x = MaxPool2D(pool_size=(2, 2))(x)
# PEPX1_Conv1x1
p_1_y = Conv2D(256, (1, 1),
padding='same',
activation='relu',
name='PEPX1_Conv')(x)
# Stage1结构
y_1_1 = PEPXModel(x, 256, 'PEPX1.1')
y_1_2 = PEPXModel(add([y_1_1, p_1_y]), 256, 'PEPX1.2')
y_1_3 = PEPXModel(add([y_1_1, y_1_2, p_1_y]), 256, 'PEPX1.3')
# PEPX2_Conv1x1
p_2_y = Conv2D(512, (1, 1),
padding='same',
activation='relu',
name='PEPX2_Conv')(add([p_1_y, y_1_1, y_1_2, y_1_3]))
p_2_y = MaxPool2D(pool_size=(2, 2))(p_2_y)
# Stage2结构
y_2_1 = PEPXModel(add([y_1_3, y_1_2, y_1_1, p_1_y]), 512, 'PEPX2.1')
y_2_1 = MaxPool2D(pool_size=(2, 2))(y_2_1)
y_2_2 = PEPXModel(add([y_2_1, p_2_y]), 512, 'PEPX2.2')
y_2_3 = PEPXModel(add([y_2_1, y_2_2, p_2_y]), 512, 'PEPX2.3')
y_2_4 = PEPXModel(add([y_2_1, y_2_2, y_2_3, p_2_y]), 512, 'PEPX2.4')
# PEPX3_Conv1x1
p_3_y = Conv2D(1024, (1, 1),
padding='same',
activation='relu',
name='PEPX3_Conv')(add([p_2_y, y_2_1, y_2_2, y_2_3, y_2_4]))
p_3_y = MaxPool2D(pool_size=(2, 2))(p_3_y)
# Stage3结构
y_3_1 = PEPXModel(add([y_2_1, y_2_2, y_2_3, y_2_4, p_2_y]), 1024,
'PEPX3.1')
y_3_1 = MaxPool2D(pool_size=(2, 2))(y_3_1)
y_3_2 = PEPXModel(y_3_1, 1024, 'PEPX3.2')
y_3_3 = PEPXModel(add([y_3_1, y_3_2]), 1024, 'PEPX3.3')
y_3_4 = PEPXModel(add([y_3_1, y_3_2, y_3_3]), 1024, 'PEPX3.4')
y_3_5 = PEPXModel(add([y_3_1, y_3_2, y_3_3, y_3_4]), 1024, 'PEPX3.5')
y_3_6 = PEPXModel(add([y_3_1, y_3_2, y_3_3, y_3_4, y_3_5]), 1024,
'PEPX3.6')
# PEPX4_Conv1x1
p_4_y = Conv2D(2048, (1, 1),
padding='same',
activation='relu',
name='PEPX4_Conv1')(add(
[p_3_y, y_3_1, y_3_2, y_3_3, y_3_4, y_3_5, y_3_6]))
p_4_y = MaxPool2D(pool_size=(2, 2))(p_4_y)
# Stage4结构
y_4_1 = PEPXModel(add([y_3_1, y_3_2, y_3_3, y_3_4, y_3_5, y_3_6, p_3_y]),
2048, 'PEPX4.1')
y_4_1 = MaxPool2D(pool_size=(2, 2))(y_4_1)
y_4_2 = PEPXModel(add([y_4_1, p_4_y]), 2048, 'PEPX4.2')
y_4_3 = PEPXModel(add([y_4_1, y_4_2, p_4_y]), 2048, 'PEPX4.3')
# FC
fla = Flatten()(add([y_4_1, y_4_2, y_4_3, p_4_y]))
d1 = Dense(1024, activation='relu')(fla) #bylo 1024 ale leciał OOM
d2 = Dense(256, activation='relu')(d1)
output = Dense(3, activation='softmax')(d2)
return keras.models.Model(input, output)
def get_cifar_backbone(embeded_dim, model_type="mlp", norm_type="bn", acti_type="relu"):
""" For `cifar10` and `cifar100` dataset
Args:
embeded_dim (int): 10 or 100
Returns:
encoder: (32, 32, 3) -> (1, 1, 10) or (1, 1, 100)
decoder: (1, 1, 10) or (1, 1, 100) -> (32, 32, 3)
"""
input_shape = (32, 32, 3)
init = "he_normal"
if model_type == "mlp":
return get_mlp_backbones(input_shape, embeded_dim, norm_type, acti_type, kernel_initializer=init)
elif model_type == "allconv":
acti = acti_type
encoder_input = Input(shape=input_shape)
encoder_layers = []
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=32, kernel_size=5, strides=2,
padding="same", kernel_initializer=init)) # 16
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=64, kernel_size=5, strides=2,
padding="same", kernel_initializer=init)) # 8
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=128, kernel_size=3, strides=2,
padding="valid", kernel_initializer=init)) # 3
encoder_layers.append(Conv2D(embeded_dim, 3, activation=None, kernel_initializer=init)) # 1
encoder_output = encoder_input
for layer in encoder_layers:
encoder_output = layer(encoder_output)
decoder_input = Input(shape=(1, 1, embeded_dim))
decoder_layers = []
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=128, kernel_size=3,
kernel_initializer=init)) # 3
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=64, kernel_size=3, output_padding=1,
strides=2, padding="valid", kernel_initializer=init)) # 8
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=32, kernel_size=5,
strides=2, padding="same", kernel_initializer=init)) # 16
decoder_layers.append(Conv2DTranspose(input_shape[-1], 5, strides=2, padding="same",
activation="sigmoid", kernel_initializer=init)) # 32
decoder_output = decoder_input
for layer in decoder_layers:
decoder_output = layer(decoder_output)
encoder = Model(inputs=encoder_input, outputs=encoder_output, name="allconvbn_encoder")
decoder = Model(inputs=decoder_input, outputs=decoder_output, name="allconvbn_decoder")
return encoder, decoder
elif model_type == "conv":
# 与原文cifar10_2保持一致的encoder结构
acti = acti_type
encoder_input = Input(shape=input_shape)
encoder_layers = []
encoder_layers.extend(n_conv_norm(3, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 26,26,64
encoder_layers.append(MaxPool2D(2, 2, padding="valid")) # 13
encoder_layers.extend(n_conv_norm(3, norm_type, acti,
filters=128, kernel_size=3,
padding="valid", kernel_initializer=init)) # 7,7,128
encoder_layers.append(MaxPool2D(2, 2, padding="valid")) # 3,3,128
encoder_layers.extend(n_conv_norm(1, norm_type, acti,
filters=embeded_dim, kernel_size=1,
padding="valid", kernel_initializer=init)) # 3,3,10
encoder_layers.append(AveragePooling2D(3))
encoder_output = encoder_input
for layer in encoder_layers:
encoder_output = layer(encoder_output)
decoder_input = Input(shape=(1, 1, embeded_dim))
decoder_layers = []
decoder_layers.append(UpSampling2D(3)) # 3
decoder_layers.extend(n_deconv_norm(1, norm_type, acti,
filters=128, kernel_size=1,
padding="valid", kernel_initializer=init)) # 3
decoder_layers.extend(n_deconv_norm(1, norm_type, acti,
filters=128, kernel_size=3, strides=2,
padding="valid", kernel_initializer=init)) # 7
decoder_layers.extend(n_deconv_norm(2, norm_type, acti,
filters=128, kernel_size=3,
padding="valid", kernel_initializer=init)) # 11
decoder_layers.extend(n_deconv_norm(1, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 13
decoder_layers.append(UpSampling2D(2))
decoder_layers.extend(n_deconv_norm(2, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 30
decoder_layers.append(Conv2DTranspose(3, kernel_size=3, kernel_initializer=init)) # 32, (no bn)
decoder_output = decoder_input
for layer in decoder_layers:
decoder_output = layer(decoder_output)
encoder = Model(inputs=encoder_input, outputs=encoder_output, name="conv_encoder")
decoder = Model(inputs=decoder_input, outputs=decoder_output, name="conv_encoder")
return encoder, decoder
def get_mnist_backbone(embeded_dim, model_type="mlp", norm_type="bn", acti_type="relu"):
""" For `mnist` and `fashion-mnist` dataset
Args:
embeded_dim (int): 10
Returns:
encoder: (28, 28, 1) -> (1, 1, 10)
decoder: (1, 1, 10) -> (28, 28, 1)
"""
input_shape = (28, 28, 1)
init = "he_normal"
if model_type == "mlp":
return get_mlp_backbones(input_shape, embeded_dim, norm_type, acti_type, kernel_initializer=init)
elif model_type == "conv":
# 与原论文略有不同
acti = acti_type
encoder_input = Input(shape=input_shape)
encoder_layers = []
encoder_layers.extend(n_conv_norm(3, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 22,22,64
encoder_layers.append(MaxPool2D(2, 2, padding="valid")) # 11
encoder_layers.extend(n_conv_norm(4, norm_type, acti,
filters=128, kernel_size=3,
padding="valid", kernel_initializer=init)) # 3,3,128
# encoder_layers.append(MaxPool2D(2, 2, padding="valid"))
encoder_layers.extend(n_conv_norm(1, norm_type, acti,
filters=embeded_dim, kernel_size=1,
padding="valid", kernel_initializer=init)) # 3,3,10
# encoder_layers.append(AveragePooling2D(2)) # 1, 1, 10
encoder_layers.append(AveragePooling2D(3)) # 1, 1, 10
encoder_output = encoder_input
for layer in encoder_layers:
encoder_output = layer(encoder_output)
decoder_input = Input(shape=(1, 1, embeded_dim))
decoder_layers = []
decoder_layers.append(UpSampling2D(3)) # 3
decoder_layers.extend(n_deconv_norm(1, norm_type, acti,
filters=128, kernel_size=1,
padding="valid", kernel_initializer=init)) # 3
decoder_layers.extend(n_deconv_norm(3, norm_type, acti,
filters=128, kernel_size=3,
padding="valid", kernel_initializer=init)) # 9
decoder_layers.extend(n_deconv_norm(1, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 11
decoder_layers.append(UpSampling2D(2))
decoder_layers.extend(n_deconv_norm(2, norm_type, acti,
filters=64, kernel_size=3,
padding="valid", kernel_initializer=init)) # 26
decoder_layers.append(Conv2DTranspose(1, 3, kernel_initializer=init)) # 28, (no bn)
decoder_output = decoder_input
for layer in decoder_layers:
decoder_output = layer(decoder_output)
encoder = Model(inputs=encoder_input, outputs=encoder_output, name="conv_encoder")
decoder = Model(inputs=decoder_input, outputs=decoder_output, name="conv_encoder")
return encoder, decoder
elif model_type == "allconv":
acti = acti_type
encoder_input = Input(shape=input_shape)
encoder_layers = []
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=32, kernel_size=5, strides=2,
padding="same", kernel_initializer=init)) # 14
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=64, kernel_size=5, strides=2,
padding="same", kernel_initializer=init)) # 7
encoder_layers.extend(n_conv_norm(1, norm_type, acti, filters=128, kernel_size=3, strides=2,
padding="valid", kernel_initializer=init)) # 3
encoder_layers.append(Conv2D(embeded_dim, 3, activation="relu", kernel_initializer=init)) # 1
encoder_output = encoder_input
for layer in encoder_layers:
encoder_output = layer(encoder_output)
decoder_input = Input(shape=(1, 1, embeded_dim))
decoder_layers = []
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=128, kernel_size=3,
kernel_initializer=init)) # 3
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=64, kernel_size=3,
strides=2, padding="valid", kernel_initializer=init)) # 7
decoder_layers.extend(n_deconv_norm(1, norm_type, acti, filters=32, kernel_size=5,
strides=2, padding="same", kernel_initializer=init)) # 14
decoder_layers.append(Conv2DTranspose(input_shape[-1], 5, strides=2, padding="same",
activation="relu", kernel_initializer=init)) # 28
decoder_output = decoder_input
for layer in decoder_layers:
decoder_output = layer(decoder_output)
encoder = Model(inputs=encoder_input, outputs=encoder_output, name="allconvbn_encoder")
decoder = Model(inputs=decoder_input, outputs=decoder_output, name="allconvbn_decoder")
return encoder, decoder
else:
print('Not defined model: ', model_type)
exit(0)
def build_model(img_shape: Tuple[int, int, int], num_classes: int,
optimizer: tf.keras.optimizers.Optimizer, learning_rate: float,
filter_block1: int, kernel_size_block1: int,
filter_block2: int, kernel_size_block2: int,
filter_block3: int, kernel_size_block3: int,
dense_layer_size: int,
kernel_initializer: tf.keras.initializers.Initializer,
activation_cls: tf.keras.layers.Activation) -> Model:
input_img = Input(shape=img_shape)
x = Conv2D(filters=filter_block1,
kernel_size=kernel_size_block1,
padding="same",
kernel_initializer=kernel_initializer)(input_img)
x = activation_cls(x)
x = Conv2D(filters=filter_block1,
kernel_size=kernel_size_block1,
padding="same",
kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
x = Conv2D(filters=filter_block2,
kernel_size=kernel_size_block2,
padding="same",
kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = Conv2D(filters=filter_block2,
kernel_size=kernel_size_block2,
padding="same",
kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
x = Conv2D(filters=filter_block3,
kernel_size=kernel_size_block3,
padding="same",
kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = Conv2D(filters=filter_block3,
kernel_size=kernel_size_block3,
padding="same",
kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
x = Flatten()(x)
x = Dense(units=dense_layer_size, kernel_initializer=kernel_initializer)(x)
x = activation_cls(x)
x = Dense(units=num_classes, kernel_initializer=kernel_initializer)(x)
y_pred = Activation("softmax")(x)
model = Model(inputs=[input_img], outputs=[y_pred])
opt = optimizer(learning_rate=learning_rate)
model.compile(loss="categorical_crossentropy",
optimizer=opt,
metrics=["accuracy"])
return model
# Part 1 - Building the CNN from tensorflow.keras import optimizers from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Conv2D, MaxPool2D, Flatten, Dropout # Initialing the CNN classifier = Sequential() # Step 1 - Convolutio Layer classifier.add(Conv2D(32, 3, 3, input_shape=(64, 64, 3), activation='relu')) #step 2 - Pooling classifier.add(MaxPool2D(pool_size=(2, 2))) # Adding second convolution layer classifier.add(Conv2D(32, 3, 3, activation='relu')) classifier.add(MaxPool2D(pool_size=(2, 2))) #Adding 3rd Concolution Layer classifier.add(Conv2D(64, 3, 3, activation='relu')) classifier.add(MaxPool2D(pool_size=(2, 2))) #Step 3 - Flattening classifier.add(Flatten()) #Step 4 - Full Connection classifier.add(Dense(256, activation='relu')) classifier.add(Dropout(0.5)) classifier.add(Dense(26, activation='softmax')) #Compiling The CNN
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
plotImages(imgs)
print(labels)
model = Sequential([
Conv2D(filters=32,
kernel_size=(3, 3),
activation='relu',
padding='same',
input_shape=(224, 224, 3)),
MaxPool2D(pool_size=(2, 2), strides=2),
Conv2D(filters=64, kernel_size=(3, 3), activation='relu', padding='same'),
MaxPool2D(pool_size=(2, 2), strides=2),
Flatten(),
Dense(units=2, activation='softmax')
])
model.sumary()
model.compile(
optimizer=Adam(learning_rate=0.0001),
loss="categorical_crossentropy",
metrics=['accuracy'],
)
model.fit(x=train_batches, validation_data=valid_batches, epochs=10, verbose=2)
encoder = OneHotEncoder()
y_train = encoder.fit_transform(y_train.reshape(-1, 1)).toarray()
y_test = encoder.fit_transform(y_test.reshape(-1, 1)).toarray()
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPool2D, Dense, Flatten, Dropout
model = Sequential()
model.add(
Conv2D(filters=100,
kernel_size=(2, 2),
padding='same',
strides=1,
input_shape=(28, 28, 1)))
model.add(MaxPool2D(pool_size=2))
model.add(Dropout(0.2))
# model.add(Conv2D(???????))
model.add(Flatten())
model.add(Dense(30))
model.add(Dense(10))
model.add(Dense(50))
model.add(Dense(10, activation='softmax'))
##완성하기
#지표는 acc 0.985
#x_test 10개 y_pred 10개 출력
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
modelpath = '../data/modelcheckpoint/k45_mnist_{epoch:02d}-{val_loss:.4f}.hdf5'
cp = ModelCheckpoint(filepath=modelpath,
from tensorflow.keras.layers import Input, Conv2D, MaxPool2D, UpSampling2D
from tensorflow.keras.models import Model
from tensorflow.keras.preprocessing.image import img_to_array, load_img
import matplotlib.pyplot as plt
from tensorflow.keras.callbacks import TensorBoard
IMAGE_SIZE = 256
CHANNEL = 3
test_dir = "C:/Users/zenkori/Documents/ObjectTracking/data/val/"
train_dir = "C:/Users/zenkori/Documents/ObjectTracking/data/train/"
print(test_dir)
input_img = Input(shape=(IMAGE_SIZE, IMAGE_SIZE,
CHANNEL)) # https://qiita.com/haru1977/items
# model define
x = Conv2D(28, (3, 3), activation='relu', padding='same')(input_img)
x = MaxPool2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = MaxPool2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
encoded = MaxPool2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)
x = UpSampling2D((2, 2))(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(28, (3, 3), activation='relu', padding='same')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(CHANNEL, (3, 3), padding='same')(x)
#
autoencoder = Model(input_img, decoded)
autoencoder.compile(optimizer='adam', loss='mean_squared_error')
autoencoder.summary()
def model_fn(
optimizer,
learning_rate,
filter_block1,
kernel_size_block1,
filter_block2,
kernel_size_block2,
filter_block3,
kernel_size_block3,
dense_layer_size,
kernel_initializer,
bias_initializer,
activation_str,
dropout_rate,
use_bn,
):
# Input
input_img = Input(shape=x_train.shape[1:])
# Conv Block 1
x = Conv2D(
filters=filter_block1,
kernel_size=kernel_size_block1,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(input_img)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = Conv2D(
filters=filter_block1,
kernel_size=kernel_size_block1,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = MaxPool2D()(x)
# Conv Block 2
x = Conv2D(
filters=filter_block2,
kernel_size=kernel_size_block2,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = Conv2D(
filters=filter_block2,
kernel_size=kernel_size_block2,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = MaxPool2D()(x)
# Conv Block 3
x = Conv2D(
filters=filter_block3,
kernel_size=kernel_size_block3,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = Conv2D(
filters=filter_block3,
kernel_size=kernel_size_block3,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = MaxPool2D()(x)
# Conv Block 3
x = Conv2D(
filters=filter_block3,
kernel_size=kernel_size_block3,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = Conv2D(
filters=filter_block3,
kernel_size=kernel_size_block3,
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer,
)(x)
if use_bn:
x = BatchNormalization()(x)
if dropout_rate > 0.0:
x = Dropout(rate=dropout_rate)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = MaxPool2D()(x)
# Dense Part
x = Flatten()(x)
x = Dense(units=dense_layer_size)(x)
if activation_str == "LeakyReLU":
x = LeakyReLU()(x)
else:
x = Activation(activation_str)(x)
x = Dense(units=num_classes)(x)
y_pred = Activation("softmax")(x)
# Build the model
model = Model(inputs=[input_img], outputs=[y_pred])
opt = optimizer(learning_rate=learning_rate)
model.compile(loss="categorical_crossentropy",
optimizer=opt,
metrics=["accuracy"])
return model
def __init__(self, learning_rate=0.003):
global_filter_size1 = 16
global_filter_size2 = 32
global_filter_size3 = 32
global_filter_partial_final = 10
global_filter_final = 10
x_input = tf.keras.Input(shape=(512, 512, 3), name="x_input_node")
first = Conv2D(kernel_size=(3, 3),
filters=global_filter_size1,
padding='same',
use_bias=False)(x_input) #512
first = BatchNormalization()(first)
first = ReLU(max_value=6)(first)
first = self.residual_layer(first,
filters=global_filter_size1,
dilation=1,
kernel_size=3) # 512
#first = self.residual_layer(first, filters=global_filter_size1, dilation=1, kernel_size=3)
#first = self.residual_layer(first, filters=global_filter_size1, dilation=1, kernel_size=3)
second = MaxPool2D(pool_size=2, strides=2)(first)
second = self.residual_layer(second,
filters=global_filter_size1,
dilation=1,
kernel_size=3) #256
#second = self.residual_layer(second, filters=global_filter_size1, dilation=1, kernel_size=3)
#second = self.residual_layer(second, filters=global_filter_size1, dilation=1, kernel_size=3)
third = MaxPool2D(pool_size=2, strides=2)(second)
third = self.residual_layer(third,
filters=global_filter_size1,
dilation=1,
kernel_size=3) #128
#third = self.residual_layer(third, filters=global_filter_size1, dilation=1, kernel_size=3)
#third = self.residual_layer(third, filters=global_filter_size1, dilation=1, kernel_size=3)
x1 = tf.image.resize(x_input, size=(256, 256))
x1 = Conv2D(kernel_size=(3, 3),
filters=global_filter_size2,
padding='same',
use_bias=False)(x1) #256
x1 = BatchNormalization()(x1)
x1 = ReLU(max_value=6)(x1)
#x1_1 = self.residual_layer(x1, filters=global_filter_size2, dilation=1, kernel_size=3)
#x1_2 = self.residual_layer(x1, filters=global_filter_size2, dilation=2, kernel_size=3)
x1_3 = self.residual_layer(x1,
filters=global_filter_size2,
dilation=2,
kernel_size=5)
x1_4 = self.residual_layer(x1,
filters=global_filter_size2,
dilation=2,
kernel_size=7)
x1_final = tf.concat([x1_3, x1_4, second], -1) # 256
x1_final = Conv2D(kernel_size=(3, 3),
filters=global_filter_partial_final,
padding='same',
use_bias=False)(x1_final) #256
x1_final = BatchNormalization()(x1_final)
x1_final = ReLU(max_value=6)(x1_final)
x2 = tf.image.resize(x_input, size=(128, 128))
x2 = Conv2D(kernel_size=(3, 3),
filters=global_filter_size3,
padding='same',
use_bias=False)(x2) #128
x2 = BatchNormalization()(x2)
x2 = ReLU(max_value=6)(x2)
x2_1 = self.residual_layer(x2,
filters=global_filter_size3,
dilation=2,
kernel_size=5)
#x2_2 = self.residual_layer(x2, filters=global_filter_size3, dilation=1, kernel_size=3)
x2_3 = self.residual_layer(x2,
filters=global_filter_size3,
dilation=2,
kernel_size=3)
x2_final = tf.concat([x2_1, x2_3, third], -1) # 128
x2_final = Conv2D(kernel_size=(3, 3),
filters=global_filter_partial_final,
padding='same',
use_bias=False)(x2_final) #128
x2_final = BatchNormalization()(x2_final)
x2_final = ReLU(max_value=6)(x2_final)
upsample1 = UpSampling2D(size=(2, 2))(x1_final)
upsample2 = UpSampling2D(size=(4, 4))(x2_final)
total_final = tf.concat([upsample1, upsample2, first], -1)
total_final = Conv2D(kernel_size=(3, 3),
filters=global_filter_final,
padding='same',
use_bias=False)(total_final)
total_final = BatchNormalization()(total_final)
total_final = ReLU(max_value=6)(total_final)
#total_final = Conv2D(kernel_size=(3, 3), filters=5, padding='same', use_bias=False)(total_final)
#total_final = BatchNormalization()(total_final)
#total_final = swish(x=total_final)
total_final = Conv2D(kernel_size=(3, 3),
filters=1,
padding='same',
use_bias=False)(total_final)
#total_final = BatchNormalization()(total_final)
output = tf.sigmoid(total_final, name='output')
self.model = Model(inputs=x_input, outputs=output)
#self.optimizer = tfa.optimizers.RectifiedAdam(learning_rate=learning_rate, warmup_proportion=0.125, total_steps=40, min_lr=0.001)
#self.optimizer = tfa.optimizers.RectifiedAdam(learning_rate=learning_rate, warmup_proportion=0, total_steps=200, min_lr=1e-4)
self.optimizer = tf.optimizers.Adam(learning_rate=learning_rate)
self.model.summary()
def train(self):
model = Sequential()
model.add(
Conv2D(filters=96,
kernel_size=(11, 11),
strides=(4, 4),
activation='relu',
input_shape=(277, 277, 3)))
model.add(BatchNormalization())
model.add(MaxPool2D(pool_size=(3, 3), strides=(2, 2)))
model.add(
Conv2D(filters=256,
kernel_size=(5, 5),
strides=(1, 1),
activation='relu',
padding='same'))
model.add(BatchNormalization())
model.add(MaxPool2D(pool_size=(3, 3), strides=(2, 2)))
model.add(
Conv2D(filters=384,
kernel_size=(3, 3),
strides=(1, 1),
activation='relu',
padding='same'))
model.add(BatchNormalization())
model.add(
Conv2D(filters=384,
kernel_size=(1, 1),
strides=(1, 1),
activation='relu',
padding='same'))
model.add(BatchNormalization())
model.add(
Conv2D(filters=256,
kernel_size=(1, 1),
strides=(1, 1),
activation='relu',
padding='same'))
model.add(BatchNormalization())
model.add(MaxPool2D(pool_size=(3, 3), strides=(2, 2)))
model.add(Flatten())
model.add(Dense(4096, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(4096, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(self.train_gen.num_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer=SGD(lr=0.001),
metrics=['accuracy'])
model.summary()
callbacks = [
ModelCheckpoint(
filepath="models/saved/AlexNet/model_epoch_{epoch}",
save_best_only=True,
monitor="val_accuracy",
verbose=1)
]
hist = model.fit(
self.train_gen,
steps_per_epoch=self.train_gen.samples // self.batch_size,
validation_data=self.test_gen,
validation_steps=self.test_gen.samples // self.batch_size,
epochs=self.epochs,
callbacks=callbacks)
plt.plot(hist.history["accuracy"])
plt.plot(hist.history['val_accuracy'])
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title("models accuracy")
plt.ylabel("Accuracy")
plt.xlabel("Epoch")
plt.legend(
["Accuracy", "Validation Accuracy", "loss", "Validation Loss"])
plt.show()
return model
def BisenetV2(include_top=True,
input_tensor=None,
input_shape=(224, 224, 3),
weights=None
):
if K.backend() != 'tensorflow':
raise RuntimeError('Only tensorflow supported for now')
name = "bisenetv2"
input_shape = _obtain_input_shape(input_shape, default_size=224, min_size=28, require_flatten=include_top,
data_format=K.image_data_format())
if input_tensor is None:
img_input = Input(shape=input_shape)
else:
if not K.is_keras_tensor(input_tensor):
img_input = Input(tensor=input_tensor, shape=input_shape)
else:
img_input = input_tensor
img_input1 = Conv2D(16, kernel_size=(3, 3), strides=2, padding="same", use_bias=False, name="stem_block/conv_block_1")(img_input)
img_input1 = BatchNormalization(axis=-1, name="stem_block/conv_block_1/bn_1")(img_input1)
img_input1 = Activation(activation="relu", name="stem_block/conv_block_1/activate_1")(img_input1)
branch_left_output = Conv2D(int(16/2), kernel_size=(1, 1), strides=1, padding="same", use_bias=False, name="stem_block/downsample_branch_left/1x1_conv_block")(img_input1)
branch_left_output = BatchNormalization(axis=-1, name="stem_block/downsample_branch_left/1x1_conv_block/bn_1")(branch_left_output)
branch_left_output = Activation(activation="relu", name="stem_block/downsample_branch_left/1x1_conv_block/activate_1")(branch_left_output)
branch_left_output = Conv2D(16, kernel_size=(3, 3), strides=2, padding="same", use_bias=False,
name="stem_block/downsample_branch_left/3x3_conv_block")(branch_left_output)
branch_left_output = BatchNormalization(axis=-1, name="stem_block/downsample_branch_left/3x3_conv_block/bn_1")(branch_left_output)
branch_left_output = Activation(activation="relu", name="stem_block/downsample_branch_left/3x3_conv_block/activate_1")(branch_left_output)
branch_right_output = MaxPool2D(pool_size=(3, 3), strides=2, padding='same', name="stem_block/downsample_branch_right/maxpooling_block")(img_input1)
stem_result = Concatenate(axis=-1, name="stem_block/concate_features")([branch_left_output, branch_right_output])
stem_result = Conv2D(16, kernel_size=(3, 3), strides=1, padding="same", use_bias=False, name="stem_block/final_conv_block")(stem_result)
stem_result = BatchNormalization(axis=-1, name="stem_block/final_conv_block/bn_1")(stem_result)
stem_result = Activation(activation="relu", name="stem_block/final_conv_block/activate_1")(stem_result)
# k_reduce_mean = Lambda(lambda x: tf.reduce_mean(x, axis=[1, 2], keepdims=True, name='global_avg_pooling'))
# embedding_result=k_reduce_mean(stem_result)
# embedding_result = K.mean(stem_result, axis=[1, 2], keepdims=True)
embedding_result = KerasReduceMean(axis=(1, 2), keep_dim=True, name="global_avg_pooling")(stem_result)
embedding_result = BatchNormalization(axis=-1, name="context_embedding_block/bn")(embedding_result)
output_channels = stem_result.get_shape().as_list()[-1]
embedding_result = Conv2D(output_channels, kernel_size=(1, 1), strides=1, padding="same", use_bias=False,
name="context_embedding_block/conv_block_1")(embedding_result)
embedding_result = BatchNormalization(axis=-1, name="context_embedding_block/conv_block_1/bn_1")(embedding_result)
embedding_result = Activation(activation="relu", name="context_embedding_block/conv_block_1/activate_1")(embedding_result)
embedding_result = Add(name="context_embedding_block/fused_features")([embedding_result, stem_result])
embedding_result = Conv2D(output_channels, kernel_size=(3, 3), strides=1, padding="same", use_bias=False, name="context_embedding_block/final_conv_block")(embedding_result)
output_channels = embedding_result.get_shape().as_list()[-1]
gather_expansion_result = Conv2D(output_channels, kernel_size=(3, 3), strides=1, padding="same", use_bias=False,
name="ge_block_with_stride_1/stride_equal_one_module/3x3_conv_block")(embedding_result)
gather_expansion_result = BatchNormalization(axis=-1, name="ge_block_with_stride_1/stride_equal_one_module/3x3_conv_block/bn_1")(gather_expansion_result)
gather_expansion_result = Activation(activation="relu", name="ge_block_with_stride_1/stride_equal_one_module/3x3_conv_block/activate_1")(gather_expansion_result)
gather_expansion_result = DepthwiseConv2D(kernel_size=3, strides=1, depth_multiplier=6, padding='same',
name="ge_block_with_stride_1/stride_equal_one_module/depthwise_conv_block")(gather_expansion_result)
gather_expansion_result = BatchNormalization(axis=-1, name="ge_block_with_stride_1/stride_equal_one_module/dw_bn")(gather_expansion_result)
gather_expansion_result = Conv2D(output_channels, kernel_size=(1, 1), strides=1, padding="same", use_bias=False,
name="ge_block_with_stride_1/stride_equal_one_module/1x1_conv_block")(gather_expansion_result)
gather_expansion_result = Add(name="ge_block_with_stride_1/stride_equal_one_module/fused_features")([embedding_result, gather_expansion_result])
gather_expansion_result = Activation(activation="relu", name="ge_block_with_stride_1/stride_equal_one_module/ge_output")(gather_expansion_result)
gather_expansion_proj_result = DepthwiseConv2D(kernel_size=3, depth_multiplier=1, strides=2, padding="same",
name="ge_block_with_stride_2/stride_equal_two_module/input_project_dw_conv_block")(gather_expansion_result)
gather_expansion_proj_result = BatchNormalization(axis=-1, name="ge_block_with_stride_2/stride_equal_two_module/input_project_bn")(gather_expansion_proj_result)
gather_expansion_proj_result = Conv2D(128, kernel_size=(1, 1), strides=1, padding="same", use_bias=False, activation=None)(gather_expansion_proj_result)
input_tensor_channels = gather_expansion_result.get_shape().as_list()[-1]
gather_expansion_stride2_result = Conv2D(input_tensor_channels, kernel_size=(3, 3), strides=1, padding="same",
use_bias=False, name="ge_block_with_stride_2/stride_equal_two_module/3x3_conv_block")(gather_expansion_result)
gather_expansion_stride2_result = BatchNormalization(axis=-1, name="ge_block_with_stride_2/stride_equal_two_module/3x3_conv_block/bn_1")(gather_expansion_stride2_result)
gather_expansion_stride2_result = Activation(activation="relu", name="ge_block_with_stride_2/stride_equal_two_module/3x3_conv_block/activate_1")(gather_expansion_stride2_result)
gather_expansion_stride2_result = DepthwiseConv2D(kernel_size=3, depth_multiplier=6, strides=2, padding="same",
name="ge_block_with_stride_2/stride_equal_two_module/depthwise_conv_block_1")(gather_expansion_stride2_result)
gather_expansion_stride2_result = BatchNormalization(axis=-1, name="ge_block_with_stride_2/stride_equal_two_module/dw_bn_1")(gather_expansion_stride2_result)
gather_expansion_stride2_result = DepthwiseConv2D(kernel_size=3, depth_multiplier=1, strides=1, padding="same",
name="ge_block_with_stride_2/stride_equal_two_module/depthwise_conv_block_2")(gather_expansion_stride2_result)
gather_expansion_stride2_result = BatchNormalization(axis=-1, name="ge_block_with_stride_2/stride_equal_two_module/dw_bn_2")(gather_expansion_stride2_result)
gather_expansion_stride2_result = Conv2D(128, kernel_size=(1, 1), strides=1, padding="same",
use_bias=False, activation=None, name="ge_block_with_stride_2/stride_equal_two_module/1x1_conv_block")(gather_expansion_stride2_result)
gather_expansion_total_result = Add(name="ge_block_with_stride_2/stride_equal_two_module/fused_features")([gather_expansion_proj_result, gather_expansion_stride2_result])
gather_expansion_total_result = Activation(activation="relu", name="ge_block_with_stride_2/stride_equal_two_module/ge_output")(gather_expansion_total_result)
gather_expansion_proj2_result = DepthwiseConv2D(kernel_size=3, depth_multiplier=1, strides=2, padding="same",
name="ge_block_with_stride_2_repeat/stride_equal_two_module/input_project_dw_conv_block")(gather_expansion_total_result)
gather_expansion_proj2_result = BatchNormalization(axis=-1, name="ge_block_with_stride_2_repeat/stride_equal_two_module/input_project_bn")(gather_expansion_proj2_result)
gather_expansion_proj2_result = Conv2D(128, kernel_size=(1, 1), strides=1, padding="same", use_bias=False, activation=None)(gather_expansion_proj2_result)
input_tensor_channels = gather_expansion_total_result.get_shape().as_list()[-1]
gather_expansion_stride2_result_repeat = Conv2D(input_tensor_channels, kernel_size=(3, 3), strides=1, padding="same",
use_bias=False, name="ge_block_with_stride_2_repeat/stride_equal_two_module/3x3_conv_block")(gather_expansion_total_result)
gather_expansion_stride2_result_repeat = BatchNormalization(axis=-1, name="ge_block_with_stride_2_repeat/stride_equal_two_module/3x3_conv_block/bn_1")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = Activation(activation="relu", name="ge_block_with_stride_2_repeat/stride_equal_two_module/3x3_conv_block/activate_1")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = DepthwiseConv2D(kernel_size=3, depth_multiplier=6, strides=2, padding="same",
name="ge_block_with_stride_2_repeat/stride_equal_two_module/depthwise_conv_block_1")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = BatchNormalization(axis=-1, name="ge_block_with_stride_2_repeat/stride_equal_two_module/dw_bn_1")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = DepthwiseConv2D(kernel_size=3, depth_multiplier=1, strides=1, padding="same",
name="ge_block_with_stride_2_repeat/stride_equal_two_module/depthwise_conv_block_2")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = BatchNormalization(axis=-1, name="ge_block_with_stride_2_repeat/stride_equal_two_module/dw_bn_2")(gather_expansion_stride2_result_repeat)
gather_expansion_stride2_result_repeat = Conv2D(128, kernel_size=(1, 1), strides=1, padding="same",
use_bias=False, activation=None, name="ge_block_with_stride_2_repeat/stride_equal_two_module/1x1_conv_block")(gather_expansion_stride2_result_repeat)
gather_expansion_total_result_repeat = Add(name="ge_block_with_stride_2_repeat/stride_equal_two_module/fused_features")([gather_expansion_proj2_result, gather_expansion_stride2_result_repeat])
gather_expansion_total_result_repeat = Activation(activation="relu", name="ge_block_with_stride_2_repeat/stride_equal_two_module/ge_output")(gather_expansion_total_result_repeat)
detail_input_tensor = stem_result
semantic_input_tensor = gather_expansion_total_result_repeat
output_channels = stem_result.get_shape().as_list()[-1]
detail_branch_remain = DepthwiseConv2D(kernel_size=3, strides=1, padding="same", depth_multiplier=1,
name="guided_aggregation_block/detail_branch/3x3_dw_conv_block")(detail_input_tensor)
detail_branch_remain = BatchNormalization(axis=-1, name="guided_aggregation_block/detail_branch/bn_1")(detail_branch_remain)
detail_branch_remain = Conv2D(output_channels, kernel_size=(1, 1), padding="same", strides=1, use_bias=False,
name="guided_aggregation_block/detail_branch/1x1_conv_block")(detail_branch_remain)
detail_branch_downsample = Conv2D(output_channels, kernel_size=(3, 3), strides=2, use_bias=False, activation=None,
padding="same", name="guided_aggregation_block/detail_branch/3x3_conv_block")(detail_input_tensor)
detail_branch_downsample = AveragePooling2D(pool_size=(3, 3), strides=2, padding="same", name="guided_aggregation_block/detail_branch/avg_pooling_block")(detail_branch_downsample)
semantic_branch_remain = DepthwiseConv2D(kernel_size=3, strides=1, padding="same", depth_multiplier=1,
name="guided_aggregation_block/semantic_branch/3x3_dw_conv_block")(semantic_input_tensor)
semantic_branch_remain = BatchNormalization(axis=-1, name="guided_aggregation_block/semantic_branch/bn_1")(semantic_branch_remain)
semantic_branch_remain = Conv2D(output_channels, kernel_size=(1, 1), strides=1, use_bias=False, activation=None, padding="same",
name="guided_aggregation_block/semantic_branch/1x1_conv_block")(semantic_branch_remain)
# semantic_branch_remain = sigmoid(semantic_branch_remain)
# keras_sigmoid = Lambda(lambda x: tf.nn.sigmoid(x, name="guided_aggregation_block/semantic_branch/semantic_remain_sigmoid"))
# semantic_branch_remain = keras_sigmoid(semantic_branch_remain)
semantic_branch_remain = Activation("sigmoid", name="guided_aggregation_block/semantic_branch/semantic_remain_sigmoid")(semantic_branch_remain)
semantic_branch_upsample = Conv2D(output_channels, kernel_size=(3, 3), strides=1, padding="same", use_bias=False,
activation=None, name="guided_aggregation_block/semantic_branch/3x3_conv_block")(semantic_input_tensor)
# semantic_branch_upsample = resize_images(semantic_branch_upsample, 4, 4, data_format="channels_last", interpolation='bilinear')
# upsample_bilinear0 = Lambda(lambda x: tf.image.resize_bilinear(x, size=stem_result.get_shape().as_list()[1:3],
# name="guided_aggregation_block/semantic_branch/semantic_upsample_features"))
# semantic_branch_upsample = upsample_bilinear0(semantic_branch_upsample)
semantic_branch_upsample = BilinearUpSampling2D((4, 4), name="guided_aggregation_block/semantic_branch/semantic_upsample_features")(semantic_branch_upsample)
semantic_branch_upsample = Activation("sigmoid", name="guided_aggregation_block/semantic_branch/semantic_branch_upsample_sigmoid")(semantic_branch_upsample)
# keras_sigmoid_1 = Lambda(lambda x: tf.nn.sigmoid(x, name="guided_aggregation_block/semantic_branch/semantic_branch_upsample_sigmoid"))
# semantic_branch_upsample = keras_sigmoid_1(semantic_branch_upsample)
# semantic_branch_upsample = sigmoid(semantic_branch_upsample)
guided_features_remain = Multiply(name="guided_aggregation_block/aggregation_features/guided_detail_features")([detail_branch_remain, semantic_branch_upsample])
guided_features_downsample = Multiply(name="guided_aggregation_block/aggregation_features/guided_semantic_features")([detail_branch_downsample, semantic_branch_remain])
# upsample_bilinear1 = Lambda(lambda x: tf.image.resize_bilinear(x, size=stem_result.get_shape().as_list()[1:3],
# name="guided_aggregation_block/aggregation_features/guided_upsample_features"))
#
# guided_features_upsample = upsample_bilinear1(guided_features_downsample)
guided_features_upsample = BilinearUpSampling2D((4, 4), name="guided_aggregation_block/aggregation_features/guided_upsample_features")(guided_features_downsample)
# guided_features_upsample = resize_images(guided_features_downsample, 4, 4, data_format="channels_last", interpolation='bilinear')
guided_features = Add(name="guided_aggregation_block/aggregation_features/fused_features")([guided_features_remain, guided_features_upsample])
guided_features = Conv2D(output_channels, kernel_size=(3, 3), strides=1, use_bias=False, padding="same",
name="guided_aggregation_block/aggregation_features/aggregation_feature_output")(guided_features)
guided_features = BatchNormalization(axis=-1, name="guided_aggregation_block/aggregation_features/aggregation_feature_output/bn_1")(guided_features)
guided_features = Activation(activation="relu", name="guided_aggregation_block/aggregation_features/aggregation_feature_output/activate_1")(guided_features)
# input_tensor_size = [int(tmp * 4)for tmp in guided_features.get_shape().as_list()[1:3]]
result = Conv2D(8, kernel_size=(3, 3), strides=1, use_bias=False, padding="same", name="seg_head_block/3x3_conv_block")(guided_features)
result = BatchNormalization(axis=-1, name="seg_head_block/bn_1")(result)
result = Activation("relu", name="seg_head_block/activate_1")(result)
# upsample_bilinear2 = Lambda(lambda x: tf.image.resize_bilinear(x, size=input_tensor_size, name="seg_head_block/segmentation_head_logits"))
# result = upsample_bilinear2(result)
result = BilinearUpSampling2D((4, 4), name="seg_head_block/segmentation_head_upsample")(result)
# result = resize_images(result, 4, 4, data_format="channels_last", interpolation='bilinear')
result = Conv2D(1, kernel_size=(1, 1), strides=1, use_bias=False, padding="same",
name="seg_head_block/1x1_conv_block")(result)
if input_tensor:
inputs = get_source_inputs(input_tensor)
else:
inputs = img_input
model = Model(inputs, result, name=name)
if weights:
model.load_weights(weights, by_name=True)
return model
optimizer = Adam(lr=lr)
epochs = 10
batch_size = 256
# Define the DNN
model = Sequential()
model.add(
Conv2D(filters=32,
kernel_size=3,
padding='same',
input_shape=x_train.shape[1:]))
model.add(Activation("relu"))
model.add(Conv2D(filters=32, kernel_size=3, padding='same'))
model.add(Activation("relu"))
model.add(MaxPool2D())
model.add(Conv2D(filters=64, kernel_size=5, padding='same'))
model.add(Activation("relu"))
model.add(Conv2D(filters=64, kernel_size=5, padding='same'))
model.add(Activation("relu"))
model.add(MaxPool2D())
model.add(Flatten())
model.add(Dense(units=128))
model.add(Activation("relu"))
model.add(Dense(units=num_classes))
model.add(Activation("softmax"))
# LSCNN modeling
nInput_LSTM = 3
nOutput_LSTM = 3
nStep_LSTM = 20
nHidden_LSTM = 30
nStep_CNN = 20
nFeature_CNN = 3
nChannel_CNN = 1
LSTM_x = Input(batch_shape=(None, nStep_LSTM, 3))
xLstm1 = LSTM(nHidden_LSTM, return_sequences=True)(LSTM_x)
xLstm2 = Bidirectional(LSTM(nHidden_LSTM), merge_mode='concat')(xLstm1)
xFlat_LSTM = Flatten()(xLstm2)
xConv = Conv2D(filters=30, kernel_size=3, \
strides=1, padding = 'same', activation='tanh')(tf.reshape(xFlat_LSTM,[-1,nStep_CNN,nFeature_CNN,nChannel_CNN]))
xPool = MaxPool2D(pool_size=(2, 2), strides=1, padding='same')(xConv)
xFlat_CNN = Flatten()(xPool)
Output_Serial = Dense(3, activation='linear')(xFlat_CNN)
model = Model(LSTM_x, Output_Serial)
model.compile(loss='mse', optimizer=Adam(lr=0.001))
model.fit(input_array_Serial, output_array_Serial, epochs=100)
pred_Serial = model.predict(test_array_Serial)
pred_Serial = pred_Serial[-20:, :]
ax1 = np.arange(1, len(scaled_df) + 1)
ax2 = np.arange(len(scaled_df), len(scaled_df) + len(pred_Serial))
plt.figure(figsize=(8, 3))
plt.plot(ax1, scaled_df, label='Time series', linewidth=1)
plt.plot(ax2, pred_Serial, label='Estimate')
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print(input_shape)
# 构建网络
model = Sequential()
# 第一个卷积层，32个卷积核，大小5x5，卷积模式SAME,激活函数relu,输入张量的大小
model.add(Conv2D(filters=32, kernel_size=(5, 5), padding='Same', activation='relu',
input_shape=(28, 28, 1)))
# 池化层,池化核大小2x2
model.add(MaxPool2D(pool_size=(2, 2)))
# 随机丢弃四分之一的网络连接，防止过拟合
model.add(Dropout(0.25))
# 卷积层，64个卷积核，大小5x5，卷积模式SAME,激活函数relu,输入张量的大小
model.add(Conv2D(filters=64, kernel_size=(5, 5), padding='Same', activation='relu'))
# 池化层,池化核大小2x2
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Dropout(0.25))
# 全连接层,展开操作，
model.add(Flatten())
# 添加隐藏层神经元的数量和激活函数
model.add(Dense(1000, activation='relu'))
model.add(Dropout(0.25))
# 输出层
# 十个神经元的
model.add(Dense(10, activation='softmax'))
def densenet_ml5(img_shape, n_classes, finalAct='softmax', f=32):
repetitions = 6, 12, 24 #,16
r2 = 10
def bn_rl_conv(x, f, k=1, s=1, p='same'):
x = BatchNormalization(epsilon=1.001e-5)(x)
x = ReLU()(x)
x = Conv2D(f, k, strides=s, padding=p)(x)
return x
def dense_block(tensor, r):
for _ in range(r):
x = bn_rl_conv(tensor, 4 * f)
x = bn_rl_conv(x, f, 3)
tensor = Concatenate()([tensor, x])
return tensor
def transition_block(x):
x = bn_rl_conv(x, K.int_shape(x)[-1] // 2)
#x = Dropout(0.3)(x)
x = AvgPool2D(2, strides=2, padding='same')(x)
return x
input = Input(img_shape)
x = Conv2D(64, 7, strides=2, padding='same')(input)
x = BatchNormalization(epsilon=1.001e-5)(x)
x = ReLU()(x)
x = MaxPool2D(3, strides=2, padding='same')(x)
for r in repetitions:
d = dense_block(x, r)
x = transition_block(d)
#x = GlobalAvgPool2D()(d)
"""
outputs = []
for i in range(n_classes):
print("class ", i)
d = dense_block(x, r2)
branch = transition_block(d)
branch = GlobalAvgPool2D()(d)
output = Dense(1, activation=finalAct)(branch)
outputs.append(output)
"""
outputs = []
a = transition_block(d)
for i in range(n_classes):
# v = Conv2D(128, (3, 3), activation='relu', padding='same')(a)
# v = Conv2D(128, (3, 3), activation='relu', padding='same')(v)
# v = Conv2D(128, (3, 3), activation='relu', padding='same')(v)
v = Conv2D(256, (3, 3), activation='relu', padding='same')(a)
v = Conv2D(256, (3, 3), activation='relu', padding='same')(v)
v = Conv2D(256, (3, 3), activation='relu', padding='same')(v)
v = MaxPooling2D((2, 2), strides=(2, 2))(v)
v = Conv2D(512, (3, 3), activation='relu', padding='same')(v)
v = Conv2D(512, (3, 3), activation='relu', padding='same')(v)
v = Conv2D(512, (3, 3), activation='relu', padding='same')(v)
v = MaxPooling2D((2, 2), strides=(2, 2))(v)
v = Flatten()(v)
d = Dense(4096, activation='relu')(v)
d = Dense(4096, activation='relu')(d)
d = Dense(1024, activation='relu')(d)
output = Dense(1, activation=finalAct)(d)
outputs.append(output)
outputs = Concatenate()(outputs)
#output = Dense(n_classes, activation=finalAct)(x)
model = Model(input, outputs)
return model
# Build evaluation pipeline
ds_test = ds_test.map(normalize_img, num_parallel_calls=AUTO)
ds_test = ds_test.batch(64)
ds_test = ds_test.cache()
ds_test = ds_test.prefetch(AUTO)
model = Sequential([
Conv2D(filters=32,
kernel_size=3,
strides=2,
activation=tf.nn.relu,
input_shape=[32, 32, 3],
data_format='channels_last',
name='Conv1'),
BatchNormalization(),
MaxPool2D(2, 2, name='MaxPool1'),
BatchNormalization(),
Conv2D(filters=64,
kernel_size=3,
strides=2,
activation=tf.nn.relu,
name='Conv2',
kernel_regularizer=l2(0.02)),
BatchNormalization(),
Conv2D(filters=128,
kernel_size=3,
strides=2,
activation=tf.nn.relu,
name='Conv3'),
BatchNormalization(),
Flatten(),
def vgg(input_shape, num_classes, finalAct="softmax"):
model = Sequential()
model.add(
Conv2D(input_shape=input_shape,
filters=64,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=64,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(
Conv2D(filters=128,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=128,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
padding="same",
activation="relu"))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Flatten())
model.add(Dense(units=4096, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(units=4096, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(units=num_classes, activation=finalAct))
return model
def __init__(self):
super(VGGNet, self).__init__()
# 当没有BN操作时时可以把激活直接写在卷积操作里 #
self.conv1 = Conv2D(filters=64, kernel_size=3, strides=1, padding='same')
self.bn1 = BatchNormalization()
self.act1 = Activation(activation='relu')
self.conv2 = Conv2D(filters=64, kernel_size=3, padding='same')
self.bn2 = BatchNormalization()
self.act2 = Activation('relu')
self.pool2 = MaxPool2D(2, 2, padding='same')
self.drop2 = Dropout(0.2)
self.conv3 = Conv2D(filters=128, kernel_size=3, padding='same')
self.bn3 = BatchNormalization()
self.act3 = Activation('relu')
self.conv4 = Conv2D(filters=128, kernel_size=3, padding='same')
self.bn4 = BatchNormalization()
self.act4 = Activation('relu')
self.pool4 = MaxPool2D(pool_size=2, strides=2, padding='same')
self.drop4 = Dropout(0.2)
self.conv5 = Conv2D(filters=256, kernel_size=3, padding='same')
self.bn5 = BatchNormalization()
self.act5 = Activation('relu')
self.conv6 = Conv2D(filters=256, kernel_size=3, padding='same')
self.bn6 = BatchNormalization()
self.act6 = Activation('relu')
self.conv7 = Conv2D(filters=256, kernel_size=3, padding='same')
self.bn7 = BatchNormalization()
self.act7 = Activation('relu')
self.pool7 = MaxPool2D(pool_size=2, strides=2, padding='same')
self.drop7 = Dropout(0.2)
self.conv8 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn8 = BatchNormalization()
self.act8 = Activation('relu')
self.conv9 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn9 = BatchNormalization()
self.act9 = Activation('relu')
self.conv10 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn10 = BatchNormalization()
self.act10 = Activation('relu')
self.pool10 = MaxPool2D(pool_size=2, strides=2, padding='same')
self.drop10 = Dropout(0.2)
self.conv11 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn11 = BatchNormalization()
self.act11 = Activation('relu')
self.conv12 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn12 = BatchNormalization()
self.act12 = Activation('relu')
self.conv13 = Conv2D(filters=512, kernel_size=3, padding='same')
self.bn13 = BatchNormalization()
self.act13 = Activation('relu')
self.pool13 = MaxPool2D(pool_size=2, strides=2, padding='same')
self.drop13 = Dropout(0.2)
self.flatten = Flatten()
self.dense14 = Dense(units=512, activation='relu')
self.drop14 = Dropout(0.2)
self.dense15 = Dense(units=512, activation='relu')
self.drop15 = Dropout(0.2)
self.dense16 = Dense(units=2, activation='softmax')
def densenet_ml1(img_shape, n_classes, finalAct='softmax', f=32):
# repetitions = 6, 12, 24, 16
repetitions = 6, 12, 32
r2 = 6
def bn_rl_conv(x, f, k=1, s=1, p='same'):
x = BatchNormalization()(x)
x = ReLU()(x)
x = Conv2D(f, k, strides=s, padding=p)(x)
return x
def dense_block(tensor, r):
for _ in range(r):
x = bn_rl_conv(tensor, 4 * f)
x = bn_rl_conv(x, f, 3)
tensor = Concatenate()([tensor, x])
return tensor
def transition_block(x):
x = bn_rl_conv(x, K.int_shape(x)[-1] // 2)
#x = Dropout(0.5)(x)
x = AvgPool2D(2, strides=2, padding='same')(x)
return x
input = Input(img_shape)
x = Conv2D(64, 7, strides=2, padding='same')(input)
x = BatchNormalization(epsilon=1.001e-5)(x)
x = ReLU()(x)
x = MaxPool2D(3, strides=2, padding='same')(x)
""" Testing
r = 3
d = dense_block(x, r)
x = transition_block(d)
"""
for r in repetitions:
d = dense_block(x, r)
x = transition_block(d)
outputs = []
for i in range(n_classes):
print("class ", i)
d = dense_block(x, r2)
branch = transition_block(d)
branch = GlobalAvgPool2D()(branch)
output = Dense(1, activation=finalAct)(branch)
outputs.append(output)
outputs = Concatenate()(outputs)
""" Example on Resnet V2
outputs = []
for i in range(n_classes):
output = output_layer(x,num_filters_in)
outputs.append(output)
# concate for output purpose
outputs = keras.layers.Concatenate()(outputs)
"""
""" Original
x = GlobalAvgPool2D()(d)
output = Dense(n_classes, activation=finalAct)(x)
"""
model = Model(input, outputs)
return model
def create_model(self):
inputs = Input(shape=(self.input_shape))
conv1_1 = Conv2D(filters=64,
kernel_size=3,
padding='same',
activation='relu')(inputs)
conv1_2 = Conv2D(filters=64,
kernel_size=3,
padding='same',
activation='relu')(conv1_1)
pool1 = MaxPool2D(pool_size=2, strides=2)(conv1_2)
conv2_1 = Conv2D(filters=128,
kernel_size=3,
padding='same',
activation='relu')(pool1)
conv2_2 = Conv2D(filters=128,
kernel_size=3,
padding='same',
activation='relu')(conv2_1)
pool2 = MaxPool2D(pool_size=2, strides=2)(conv2_2)
conv3_1 = Conv2D(filters=256,
kernel_size=3,
padding='same',
activation='relu')(pool2)
conv3_2 = Conv2D(filters=256,
kernel_size=3,
padding='same',
activation='relu')(conv3_1)
pool3 = MaxPool2D(pool_size=2, strides=2)(conv3_2)
conv4_1 = Conv2D(filters=512,
kernel_size=3,
padding='same',
activation='relu')(pool3)
conv4_2 = Conv2D(filters=512,
kernel_size=3,
padding='same',
activation='relu')(conv4_1)
pool4 = MaxPool2D(pool_size=2, strides=2)(conv4_2)
conv5_1 = Conv2D(filters=1024,
kernel_size=3,
padding='same',
activation='relu')(pool4)
conv5_2 = Conv2D(filters=1024,
kernel_size=3,
padding='same',
activation='relu')(conv5_1)
concated1 = concatenate([
conv4_2,
Conv2DTranspose(filters=512,
kernel_size=2,
strides=2,
padding='same')(conv5_2)
])
conv6_1 = Conv2D(filters=512,
kernel_size=3,
padding='same',
activation='relu')(concated1)
conv6_2 = Conv2D(filters=512,
kernel_size=3,
padding='same',
activation='relu')(conv6_1)
concated2 = concatenate([
conv3_2,
Conv2DTranspose(filters=256,
kernel_size=2,
strides=2,
padding='same')(conv6_2)
])
conv7_1 = Conv2D(filters=256,
kernel_size=3,
padding='same',
activation='relu')(concated2)
conv7_2 = Conv2D(filters=256,
kernel_size=3,
padding='same',
activation='relu')(conv7_1)
concated3 = concatenate([
conv2_2,
Conv2DTranspose(filters=128,
kernel_size=2,
strides=2,
padding='same')(conv7_2)
])
conv8_1 = Conv2D(filters=128,
kernel_size=3,
padding='same',
activation='relu')(concated3)
conv8_2 = Conv2D(filters=128,
kernel_size=3,
padding='same',
activation='relu')(conv8_1)
concated4 = concatenate([
conv1_2,
Conv2DTranspose(filters=64,
kernel_size=2,
strides=2,
padding='same')(conv8_2)
])
conv9_1 = Conv2D(filters=64,
kernel_size=3,
padding='same',
activation='relu')(concated4)
conv9_2 = Conv2D(filters=64,
kernel_size=3,
padding='same',
activation='relu')(conv9_1)
logits = Conv2D(filters=self.class_num,
kernel_size=1,
padding='same',
activation='softmax')(conv9_2)
model = Model(inputs=inputs, outputs=logits)
model.compile(optimizer='Adam',
loss=sparse_categorical_crossentropy,
metrics=get_metrics())
self.model = model
def resnet_block(x, filters, reps, strides):
x = projection_block(x, filters, strides)
for _ in range(reps - 1):
x = identity_block(x, filters)
return x
#Model
input = Input(shape=(224, 224, 3))
x = conv_batchnorm_relu(input, filters=64, kernel_size=7, strides=2)
x = MaxPool2D(pool_size=3, strides=2)(x)
x = resnet_block(x, filters=64, reps=3, strides=1)
x = resnet_block(x, filters=128, reps=4, strides=2)
x = resnet_block(x, filters=256, reps=6, strides=2)
x = resnet_block(x, filters=512, reps=3, strides=2)
x = GlobalAvgPool2D()(x)
output = Dense(1000, activation='softmax')(x)
model = Model(inputs=input, outputs=output)
model.summary()
from tensorflow.python.keras.utils.vis_utils import model_to_dot
from IPython.display import SVG
import pydot
import graphviz
X_test = X_test.astype('float32')
# normalizing the data to help with the training
X_train /= 255
X_test /= 255
# building a linear stack of layers with the sequential model
model = Sequential()
# convolutional layer
model.add(Conv2D(50, kernel_size=(3,3), strides=(1,1), padding='same', activation='relu', input_shape=(32, 32, 3)))
# convolutional layer
model.add(Conv2D(75, kernel_size=(3,3), strides=(1,1), padding='same', activation='relu'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.25))
model.add(Conv2D(125, kernel_size=(3,3), strides=(1,1), padding='same', activation='relu'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.25))
# flatten output of conv
model.add(Flatten())
# hidden layer
model.add(Dense(500, activation='relu'))
model.add(Dropout(0.4))
model.add(Dense(250, activation='relu'))
model.add(Dropout(0.3))
# output layer
x_train = x_train.reshape(40000, 32, 32, 3) / 225.
x_test = x_test.reshape(10000, 32, 32, 3) / 225.
x_val = x_val.reshape(10000, 32, 32, 3) / 225.
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Conv2D, MaxPool2D, Flatten
#2. 모델링
model = Sequential()
model.add(
Conv2D(filters=10,
kernel_size=(5, 5),
strides=1,
padding='same',
input_shape=(32, 32, 3)))
model.add(MaxPool2D(pool_size=(3, 3)))
model.add(Flatten())
model.add(Dense(10))
model.add(Dense(10))
model.add(Dense(1))
#3. 컴파일, 훈련
model.compile(loss='mse', optimizer='adam', metrics=['mae'])
'''
EarlyStopping
'''
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
modelpath = '../modelcheckpoint/k46_2_cifar10_{epoch:02d}-{val_loss:.4f}.hdf5'
cp = ModelCheckpoint(filepath=modelpath,
monitor='val_loss',
save_best_only=True,