def vgg_(in_shape=(227, 227, 3),
n_classes=1000,
opt='sgd',
n_stages_per_blocks=[2, 2, 3, 3, 3]):
in_layer = layers.Input(in_shape)
block1 = block(in_layer, 64, n_stages_per_blocks[0])
pool1 = layers.MaxPool2D()(block1)
block2 = block(pool1, 128, n_stages_per_blocks[1])
pool2 = layers.MaxPool2D()(block2)
block3 = block(pool2, 256, n_stages_per_blocks[2])
pool3 = layers.MaxPool2D()(block3)
block4 = block(pool3, 512, n_stages_per_blocks[3])
pool4 = layers.MaxPool2D()(block4)
block5 = block(pool4, 512, n_stages_per_blocks[4])
pool5 = layers.MaxPool2D()(block5)
flattened = layers.GlobalAvgPool2D()(pool5)
dense1 = layers.Dense(4096, activation='relu')(flattened)
dense2 = layers.Dense(4096, activation='relu')(dense1)
preds = layers.Dense(n_classes, activation='softmax')(dense2)
model = Model(in_layer, preds)
model.compile(loss="categorical_crossentropy",
optimizer=opt,
metrics=["accuracy"])
return model
def _resnet(block, blocks_num, im_width=224, im_height=224, num_classes=2, include_top=True):
# tensorflow中的tensor通道排序是NHWC
# (None, 224, 224, 3)
input_image = layers.Input(shape=(im_height, im_width, 3), dtype="float32")
x = layers.Conv2D(filters=64, kernel_size=7, strides=2,
padding="SAME", use_bias=False, name="conv1")(input_image)
x = layers.BatchNormalization(momentum=0.9, epsilon=1e-5, name="conv1/BatchNorm")(x)
x = layers.ReLU()(x)
x = layers.MaxPool2D(pool_size=3, strides=2, padding="SAME")(x)
x = _make_layer(block, x.shape[-1], 64, blocks_num[0], name="block1")(x)
x = _make_layer(block, x.shape[-1], 128, blocks_num[1], strides=2, name="block2")(x)
x = _make_layer(block, x.shape[-1], 256, blocks_num[2], strides=2, name="block3")(x)
x = _make_layer(block, x.shape[-1], 512, blocks_num[3], strides=2, name="block4")(x)
if include_top:
x = layers.GlobalAvgPool2D()(x) # pool + flatten
x = layers.Dense(num_classes, name="logits")(x)
predict = layers.Softmax()(x)
else:
predict = x
model = Model(inputs=input_image, outputs=predict)
return model
def _build_model(self):
backbone_model = self._get_backbone_model()
assert backbone_model is not None, f'backbone should be one of {list(_FEATURE_LAYERS.keys())}'
x = layers.concatenate([layers.GlobalAvgPool2D()(backbone_model.output),
layers.GlobalMaxPool2D()(backbone_model.output)])
o = layers.Dense(self.n_classes, activation='sigmoid', name='classification_output')(x)
self.model = models.Model(inputs=backbone_model.input, outputs=o)
def _cifar_resnet(input_shape,
num_classes,
normalization,
num_stacks,
mode,
name,
weight_decay=5e-4,
**kwargs):
if normalization.lower() not in ['batchnorm', 'unitization']:
raise Exception('Unknown normalization: {}'.format(normalization))
if normalization.lower() == 'batchnorm':
kwargs.pop('image_size', None)
else:
kwargs.setdefault('image_size', 32**2)
inputs = Kl.Input(input_shape, name=name + '-inputs')
tensor = inputs
tensor = get_convolutional_layer(16, (3, 3), (1, 1),
name + '-pre-conv',
weight_decay)(tensor)
tensor = ResNets._norm_relu(tensor, normalization, name + '-pre-',
**kwargs)
filters = [16] * num_stacks \
+ [32] * num_stacks \
+ [64] * num_stacks
strides = [(1, 1)] * num_stacks \
+ [(2, 2)] + [(1, 1)] * (num_stacks - 1) \
+ [(2, 2)] + [(1, 1)] * (num_stacks - 1)
Block._block_count = 0
for filter, stride in zip(filters, strides):
tensor = Block(filter=filter,
stride=stride,
mode=mode,
normalization=normalization,
weight_decay=weight_decay,
model_name=name)(tensor, **kwargs)
if 'image_size' in kwargs:
if isinstance(stride, int):
kwargs['image_size'] //= stride
else:
kwargs['image_size'] //= np.prod(stride)
tensor = ResNets._norm_relu(tensor, normalization, name + '-fin-',
**kwargs)
tensor = Kl.GlobalAvgPool2D(name=name + '-fin-pool')(tensor)
outputs = Kl.Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal',
kernel_regularizer=Kr.l2(weight_decay),
name=name + '-outputs')(tensor)
return Km.Model(inputs, outputs, name=name)
def __init__(self, classes):
super(ResNet, self).__init__()
self.conv7 = layers.Conv2D(64, 7, padding='same', name='conv7')
self.bn = layers.BatchNormalization()
self.act = layers.Activation('relu')
self.max_pool = layers.MaxPool2D((3, 3))
self.ib1 = IdentityBlock(64, 3, name='ID1')
self.ib2 = IdentityBlock(64, 3, name='ID2')
self.global_pool = layers.GlobalAvgPool2D()
self.classifier = layers.Dense(classes, activation='softmax')
def resnet50_head(features):
filters = 512
x = features
x = conv_block_5d(x,
3, [filters, filters, filters * 4],
stage=5,
block='a',
strides=(2, 2))
x = identity_block_5d(x,
3, [filters, filters, filters * 4],
stage=5,
block='b')
x = identity_block_5d(x,
3, [filters, filters, filters * 4],
stage=5,
block='c')
# 全局平均池化(batch_size,roi_num,channels)
x = layers.TimeDistributed(layers.GlobalAvgPool2D())(x)
return x
def build(self, input_shape):
self.channels = input_shape[-1]
if self.channels is None:
raise ValueError(
'Channel dimension of the inputs should be defined. Found `None`.'
)
self.avg = models.Sequential([
layers.GlobalAvgPool2D(keepdims=True),
layers.BatchNormalization(), # TODO: not fuse?
layers.Dropout(self.confidence)
])
self.qkv = SameConv(self.channels * 3, 1, use_bias=False)
self.proj = SameConv(self.channels,
1,
use_bias=False,
activation='sigmoid')
super().build(input_shape)
def inception_net(in_shape=(224, 224, 3), n_classes=1000, opt='sgd'):
in_layer = layers.Input(in_shape)
conv1 = layers.Conv2D(64, 7, strides=2, activation='relu',
padding='same')(in_layer)
pad1 = layers.ZeroPadding2D()(conv1)
pool1 = layers.MaxPool2D(3, 2)(pad1)
conv2_1 = conv1x1(64)(pool1)
conv2_2 = conv3x3(192)(conv2_1)
pad2 = layers.ZeroPadding2D()(conv2_2)
pool2 = layers.MaxPool2D(3, 2)(pad2)
inception3a = inception_module(pool2, 64, 96, 128, 16, 32, 32)
inception3b = inception_module(inception3a, 128, 128, 192, 32, 96, 64)
pad3 = layers.ZeroPadding2D()(inception3b)
pool3 = layers.MaxPool2D(3, 2)(pad3)
inception4a = inception_module(pool3, 192, 96, 208, 16, 48, 64)
inception4b = inception_module(inception4a, 160, 112, 224, 24, 64, 64)
inception4c = inception_module(inception4b, 128, 128, 256, 24, 64, 64)
inception4d = inception_module(inception4c, 112, 144, 288, 32, 48, 64)
inception4e = inception_module(inception4d, 256, 160, 320, 32, 128, 128)
pad4 = layers.ZeroPadding2D()(inception4e)
pool4 = layers.MaxPool2D(3, 2)(pad4)
aux_clf1 = aux_clf(inception4a)
aux_clf2 = aux_clf(inception4d)
inception5a = inception_module(pool4, 256, 160, 320, 32, 128, 128)
inception5b = inception_module(inception5a, 384, 192, 384, 48, 128, 128)
pad5 = layers.ZeroPadding2D()(inception5b)
pool5 = layers.MaxPool2D(3, 2)(pad5)
avg_pool = layers.GlobalAvgPool2D()(pool5)
dropout = layers.Dropout(0.4)(avg_pool)
preds = layers.Dense(1000, activation='softmax')(dropout)
model = Model(in_layer, [preds, aux_clf1, aux_clf2])
model.compile(loss="categorical_crossentropy",
optimizer=opt,
metrics=["accuracy"])
return model
def __init__(self):
super(Channel_attention, self).__init__()
self.reduction_ratio = 0.125
self.channel_axis = 1 if K.image_data_format(
) == "channels_first" else 3
self.maxpool = KL.GlobalMaxPooling2D()
self.avgpool = KL.GlobalAvgPool2D()
self.sigmoid = KL.Activation('sigmoid')
self.channel = 128
self.Dense_One = KL.Dense(units=int(self.channel *
self.reduction_ratio),
activation='relu',
kernel_initializer='he_normal',
use_bias=True,
bias_initializer='zeros')
self.Dense_Two = KL.Dense(units=int(self.channel),
activation='relu',
kernel_initializer='he_normal',
use_bias=True,
bias_initializer='zeros')
def cifar_base_net(inputs):
x = layers.Conv2D(32, (3, 3), padding='same', name='conv_1')(inputs)
x = layers.BatchNormalization(axis=3, name='bn_conv1')(x)
x = layers.Activation('relu')(x)
x = layers.Conv2D(32, (3, 3), name='conv_2')(x)
x = layers.Activation('relu')(x)
x = layers.MaxPooling2D(pool_size=(2, 2))(x)
x = layers.Dropout(0.25)(x)
x = layers.Conv2D(64, (3, 3), padding='same', name='conv_3')(x)
x = layers.Activation('relu')(x)
x = layers.Conv2D(64, (3, 3), name='conv_4')(x)
x = layers.Activation('relu')(x)
x = layers.MaxPooling2D(pool_size=(2, 2))(x)
x = layers.Dropout(0.25)(x)
x = layers.Conv2D(512, (3, 3), name='conv_5')(x)
x = layers.Activation('relu')(x)
x = layers.GlobalAvgPool2D()(x)
return x
def build(self, input_shape):
channels = input_shape[-1]
if channels is None:
raise ValueError(
'Channel dimension of the inputs should be defined. Found `None`.'
)
self.input_spec = layers.InputSpec(ndim=4, axes={-1: channels})
self.attend = models.Sequential([
layers.GlobalAvgPool2D(keepdims=True),
SameConv(channels,
1,
activation='sigmoid',
use_bias=False,
kernel_initializer='he_uniform')
])
self.conv = SameConv(self.filters,
1,
use_bias=False,
kernel_initializer='he_uniform')
super().build(input_shape)
def _post_res_blocks(in_tensor, n_classes): pool = layers.GlobalAvgPool2D()(in_tensor) preds = layers.Dense(n_classes, activation='softmax')(pool) return preds
layers.SeparableConv2D(32,
3,
activation='relu',
input_shape=(
height,
width,
channels,
)))
model.add(layers.SeparableConv2D(64, 3, activation='relu'))
model.add(layers.MaxPool2D(2))
model.add(layers.SeparableConv2D(64, 3, activation='relu'))
model.add(layers.SeparableConv2D(128, 3, activation='relu'))
model.add(layers.MaxPool2D(2))
model.add(layers.SeparableConv2D(64, 3, activation='relu'))
model.add(layers.SeparableConv2D(128, 3, activation='relu'))
model.add(layers.GlobalAvgPool2D())
model.add(layers.Dense(32, activation='relu'))
model.add(layers.Dense(num_classes, activation='softmax'))
model.summary()
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['acc'])
#%% 超参数优化问题
# 可以尝试使用Python的Hyperopt库或Hyperas库，后者将Hyperopt与Keras模型集成
#%% 模型集成(model ensembling)
# 将一组分类器结果汇集在一起，即分类器集成(ensemble the calssfiers)，可以得到
# 好于所以模型的结果，最简单的方式是取平均值，更聪明的方式是加权平均
# (采用Nelder-Mead方法优化)
# 另一种集成方式是宽且深(wide and deep)的模型类型，它集合了深度学习和浅层学习
# horizontal_flip=True, fill_mode="nearest") input_shape = (224, 224) model_input_shape = (224, 224, 3) model_input = Input(shape=model_input_shape, name = 'image') x = layers.Conv2D( filters = 32, kernel_size = (3, 3), strides=2, padding='same', activation = 'relu', use_bias=False, kernel_initializer = he_normal(seed = 426))(model_input) x = layers.BatchNormalization()(x) x = layers.Conv2D( filters = 64, kernel_size = (3, 3), strides=2, padding='same', activation = 'relu', use_bias=False,kernel_initializer = he_normal(seed = 426))(x) x = layers.BatchNormalization()(x) x = layers.Conv2D( filters = 128, kernel_size = (3, 3), strides=2, padding='same', activation = 'relu', use_bias=False,kernel_initializer = he_normal(seed = 426))(x) x = layers.BatchNormalization()(x) x = layers.Conv2D( filters = 256, kernel_size = (3, 3), strides=2, padding='same', activation = 'relu', use_bias=False,kernel_initializer = he_normal(seed = 426))(x) x = layers.BatchNormalization()(x) x = layers.Conv2D( filters = 512, kernel_size = (3, 3), strides=2, padding='same', activation = 'relu', use_bias=False,kernel_initializer = he_normal(seed = 426))(x) x = layers.BatchNormalization()(x) x = layers.GlobalAvgPool2D()(x) x = layers.GaussianNoise(0.1)(x) x = layers.Dropout(0.2)(x) x = layers.Dense(2, activation = 'softmax', name = 'top')(x) model = Model(model_input, x) model.summary() model.compile(adam(), loss = 'binary_crossentropy') """ ABC-Mart.... """ from keras.callbacks import TensorBoard, ModelCheckpoint, ReduceLROnPlateau, EarlyStopping log_dir = 'logs/oc_cnn/'
