def __init__(self):
super(Generator, self).__init__()
# z:[b, 100]->[b, 3*3*512]->[b, 3, 3, 512]->[b,64,64,3]
self.fc = layers.Dense(3 * 3 * 512) #[b,100]->[b,3*3*512]
self.conv1 = layers.Conv2DTranspose(256, 3, 3, 'valid')
self.bn1 = layers.BatchNormalization()
self.conv2 = layers.Conv2DTranspose(128, 5, 2, 'valid')
self.bn2 = layers.BatchNormalization()
self.conv3 = layers.Conv2DTranspose(3, 4, 3, 'valid')
def tf_model_full():
model = keras.Sequential([
layers.BatchNormalization(input_shape=[
40,
]),
layers.Dense(4, activation='relu'),
layers.BatchNormalization(),
layers.Dense(8, activation='relu'),
layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer=tf.optimizers.SGD(lr=0.001,
decay=1e-5,
nesterov=True,
momentum=0.9),
loss='binary_crossentropy',
batch_size=256,
metrics=['accuracy'])
return model
def __init__(self,_lambda,layer_dims,num_classes=10,kind='res'):
super(ResNet,self).__init__()
self.stem=Sequential([layers.Conv2D(64,(3,3),strides=(1,1)),layers.BatchNormalization(),layers.ReLU(),layers.MaxPool2D(pool_size=(2,2),strides=(1,1),padding='same')])
self.layer1=self.bulid_block(64,layer_dims[0],kind=kind)
self.layer2=self.bulid_block(128,layer_dims[1],stride=2,kind=kind)
self.layer3=self.bulid_block(256,layer_dims[2],stride=2,kind=kind)
self.layer4=self.bulid_block(512,layer_dims[3],stride=2,kind=kind)
self.avg_pool=layers.GlobalAveragePooling2D()
self.fcm=layers.Dense(1000,activation='relu',kernel_regularizer=regularizers.l2(_lambda))
self.fc=layers.Dense(num_classes)
def __init__(self,filter_num,stride=1,kind='res'):
super(BasicBlock,self).__init__()
self.kind=kind
if kind=='res':
self.conv1=layers.Conv2D(filter_num,(3,3),strides=stride,padding='same')
self.bn1=layers.BatchNormalization()
self.relu=layers.ReLU()
self.conv2=layers.Conv2D(filter_num,(3,3),strides=1,padding='same')
self.bn2=layers.BatchNormalization()
if stride!=1:
self.downsample=Sequential()
self.downsample.add(layers.Conv2D(filter_num,(1,1),strides=stride))
else:
self.downsample= lambda x:x
elif kind=='Dense':
self.conv1=layers.Conv2D(filter_num,(3,3),strides=stride,padding='same')
self.bn1=layers.BatchNormalization()
self.relu=layers.ReLU()
self.conv2=layers.Conv2D(filter_num,(3,3),strides=1,padding='same')
self.bn2=layers.BatchNormalization()
self.conv3=layers.Conv2D(filter_num,(3,3),strides=1,padding='same')
self.bn3=layers.BatchNormalization()
self.conv4=layers.Conv2D(filter_num,(3,3),strides=1,padding='same')
self.bn4=layers.BatchNormalization()
if stride!=1:
self.downsample13=Sequential()
self.downsample13.add(layers.Conv2D(filter_num,(1,1),strides=stride))
self.downsample14=Sequential()
self.downsample14.add(layers.Conv2D(filter_num,(1,1),strides=stride))
self.downsample15=Sequential()
self.downsample15.add(layers.Conv2D(filter_num,(1,1),strides=stride))
self.downsample=lambda x:x
else:
self.downsample= lambda x:x
def getModel():
model = keras.Sequential([
layers.Input(shape=[4,]),
layers.BatchNormalization(),
layers.Dense(16,activation='relu'),
layers.BatchNormalization(),
layers.Dense(32,activation='relu'),
layers.Dense(1,activation='sigmoid')
])
# lr=0.001, decay=1e-5, nesterov=True, momentum=0.9
model.compile(optimizer=tf.optimizers.SGD(lr=0.001, decay=1e-5, nesterov=True, momentum=0.9),
loss='binary_crossentropy',
batch_size=32,
metrics=['accuracy']
)
'''
model.compile(optimizer="rmsprop",
loss='binary_crossentropy',
metrics=['accuracy'],
)
'''
return model
def __init__(self):
super(Discriminator, self).__init__()
# [b,64,64,3]->[b, 20,20,64]
# Conv2D(filters, kernel_size, strides, padding)
# filters 输出的空间维度，即卷积中的滤波器数
# kernel_size 卷积核的大小。2个整数的整数元组/列表,指定2D卷积窗口的高度和宽度.可以是单个整数,以指定所有空间维度的相同值。
# 这里的5就是5*5
# strides 2个整数的整数或元组/列表,指定卷积沿高度和宽度的跨度.可以是单个整数,以指定所有空间维度的相同值.指定任何步幅值！= 1与指定任何dilation_rate值！= 1都不相容
self.con1 = layers.Conv2D(64, 5, 3, 'valid')
# [b,20,20,3]->[b, 20,20,128]
self.con2 = layers.Conv2D(128, 5, 3, 'valid')
self.bn2 = layers.BatchNormalization()
# [b,20,20,128]->[b, 20,20,256]
self.con3 = layers.Conv2D(256, 5, 3, 'valid')
self.bn3 = layers.BatchNormalization()
# 打平 [b, h, w, c]->[b, -1] 就保留第0维度，其余的维度相乘，如[b, 2, 2, 3]->[b, 2*2*3]
self.flatten = layers.Flatten()
# 全连接层[b, -1]->[b, 1]
self.fc = layers.Dense(1)