打印出tensorflow2.x的神经网络结构

使用tensorflow中plot_model函数，以VGG-16网络为例子，全部代码如下。

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Conv1D, MaxPool1D, LSTM, BatchNormalization, Dropout, Flatten, Reshape
from tensorflow.keras.utils import plot_model

model = Sequential()

# 1nd block
model.add(Conv1D(64, 3, activation="relu",
                 input_shape=(1, 1), padding="same"))
model.add(Conv1D(64, 3, activation="relu", padding="same"))
model.add(BatchNormalization(trainable=True))
model.add(MaxPool1D(pool_size=2, strides=1, padding='same'))

# 2nd block
model.add(Conv1D(128, 3, activation="relu", padding="same"))
model.add(Conv1D(128, 3, activation="relu", padding="same"))
model.add(BatchNormalization(trainable=True))
model.add(MaxPool1D(pool_size=2, strides=1, padding='same'))

# 3nd block
model.add(Conv1D(256, 3, activation="relu", padding="same"))
model.add(Conv1D(256, 3, activation="relu", padding="same"))
model.add(Conv1D(256, 3, activation="relu", padding="same"))
model.add(BatchNormalization(trainable=True))
model.add(MaxPool1D(pool_size=2, strides=1, padding='same'))

# 4nd block
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(BatchNormalization(trainable=True))
model.add(MaxPool1D(pool_size=2, strides=1, padding='same'))

# 5nd block
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(Conv1D(512, 3, activation="relu", padding="same"))
model.add(BatchNormalization(trainable=True))
model.add(MaxPool1D(pool_size=2, strides=1, padding='same'))

# 6nd block
model.add(LSTM(512))
model.add(BatchNormalization(trainable=True))

# 7nd block
model.add(Dense(512, activation='relu'))
model.add(Dense(128, activation='tanh'))
model.add(Dense(1, activation='linear'))

model.summary()
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam')
plot_model(model, to_file='VGG16_1D_简单测试模型_001.png', show_shapes=True)

模型结构输出为

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
conv1d (Conv1D)              (None, 1, 64)             256
_________________________________________________________________
conv1d_1 (Conv1D)            (None, 1, 64)             12352
_________________________________________________________________
batch_normalization (BatchNo (None, 1, 64)             256
_________________________________________________________________
max_pooling1d (MaxPooling1D) (None, 1, 64)             0
_________________________________________________________________
conv1d_2 (Conv1D)            (None, 1, 128)            24704
_________________________________________________________________
conv1d_3 (Conv1D)            (None, 1, 128)            49280
_________________________________________________________________
batch_normalization_1 (Batch (None, 1, 128)            512
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, 1, 128)            0
_________________________________________________________________
conv1d_4 (Conv1D)            (None, 1, 256)            98560
_________________________________________________________________
conv1d_5 (Conv1D)            (None, 1, 256)            196864
_________________________________________________________________
conv1d_6 (Conv1D)            (None, 1, 256)            196864
_________________________________________________________________
batch_normalization_2 (Batch (None, 1, 256)            1024
_________________________________________________________________
max_pooling1d_2 (MaxPooling1 (None, 1, 256)            0
_________________________________________________________________
conv1d_7 (Conv1D)            (None, 1, 512)            393728
_________________________________________________________________
conv1d_8 (Conv1D)            (None, 1, 512)            786944
_________________________________________________________________
conv1d_9 (Conv1D)            (None, 1, 512)            786944
_________________________________________________________________
batch_normalization_3 (Batch (None, 1, 512)            2048
_________________________________________________________________
max_pooling1d_3 (MaxPooling1 (None, 1, 512)            0
_________________________________________________________________
conv1d_10 (Conv1D)           (None, 1, 512)            786944
_________________________________________________________________
conv1d_11 (Conv1D)           (None, 1, 512)            786944
_________________________________________________________________
conv1d_12 (Conv1D)           (None, 1, 512)            786944
_________________________________________________________________
batch_normalization_4 (Batch (None, 1, 512)            2048
_________________________________________________________________
max_pooling1d_4 (MaxPooling1 (None, 1, 512)            0
_________________________________________________________________
lstm (LSTM)                  (None, 512)               2099200
_________________________________________________________________
batch_normalization_5 (Batch (None, 512)               2048
_________________________________________________________________
dense (Dense)                (None, 512)               262656
_________________________________________________________________
dense_1 (Dense)              (None, 128)               65664
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 129
=================================================================
Total params: 7,342,913
Trainable params: 7,338,945
Non-trainable params: 3,968
_________________________________________________________________

保存为png图片，如下。

有用请点个赞！！
本站所有文章均为原创，欢迎转载，请注明文章出处：https://blog.csdn.net/weixin_45092662。百度和各类采集站皆不可信，搜索请谨慎鉴别。技术类文章一般都有时效性，本人习惯不定期对自己的博文进行修正和更新，因此请访问出处以查看本文的最新版本。