【596】keras显示网络结构图

参考1：【推荐】怎么显示Keras的网络结构和其中的参数

参考2：【推荐】Mac BigSur：安装homebrew(国内源)+Graphviz

参考3：mac下的Graphviz安装及使用

参考4：Mac 安装 Graphviz-python

　　也可以在线实现，https://netron.app，需要 *.h5 文件，信息更智能，但是没有输入，不过用了不同颜色显示。（示例图在文末）

---

1. 安装 pydot

?

1	pip install pydot

2. 安装 pydot-ng

?

1	pip install pydot-ng

3. 安装 homebrew（国内源）

　　打开终端，输入以下代码：

?

1	/bin/zsh -c "$(curl -fsSL https://gitee.com/cunkai/HomebrewCN/raw/master/Homebrew.sh)"

　　选择序号：1

　　执行脚本：Y 

4. 安装 Graphviz

　　homebrew安装完毕后，运行 brew install graphviz即可

?

1	brew install graphviz

　　参考2可以有效解决

　　举例：

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71

from tensorflow.keras import layers   def get_model(img_size, num_classes):     # 二维变三维，元组加法相当于 concat     inputs = keras.Input(shape=img_size + (3,))       ### [First half of the network: downsampling inputs] ###       # Entry block     x = layers.Conv2D(32, 3, strides=2, padding="same")(inputs)     x = layers.BatchNormalization()(x)     x = layers.Activation("relu")(x)       previous_block_activation = x  # Set aside residual       # Blocks 1, 2, 3 are identical apart from the feature depth.     for filters in [64, 128, 256]:         x = layers.Activation("relu")(x)         x = layers.SeparableConv2D(filters, 3, padding="same")(x)         x = layers.BatchNormalization()(x)           x = layers.Activation("relu")(x)         x = layers.SeparableConv2D(filters, 3, padding="same")(x)         x = layers.BatchNormalization()(x)           x = layers.MaxPooling2D(3, strides=2, padding="same")(x)           # Project residual         residual = layers.Conv2D(filters, 1, strides=2, padding="same")(             previous_block_activation         )         x = layers.add([x, residual])  # Add back residual         previous_block_activation = x  # Set aside next residual       ### [Second half of the network: upsampling inputs] ###       for filters in [256, 128, 64, 32]:         x = layers.Activation("relu")(x)         x = layers.Conv2DTranspose(filters, 3, padding="same")(x)         x = layers.BatchNormalization()(x)           x = layers.Activation("relu")(x)         x = layers.Conv2DTranspose(filters, 3, padding="same")(x)         x = layers.BatchNormalization()(x)           x = layers.UpSampling2D(2)(x)           # Project residual         residual = layers.UpSampling2D(2)(previous_block_activation)         residual = layers.Conv2D(filters, 1, padding="same")(residual)         x = layers.add([x, residual])  # Add back residual         previous_block_activation = x  # Set aside next residual       # Add a per-pixel classification layer     outputs = layers.Conv2D(num_classes, 3, activation="softmax", padding="same")(x)       # Define the model     model = keras.Model(inputs, outputs)     return model     # Free up RAM in case the model definition cells were run multiple times keras.backend.clear_session()   # Build model model = get_model(img_size, num_classes) model.summary()   # 结构图显示 from keras.utils.vis_utils import plot_model plot_model(model, to_file='Flatten.png', show_shapes=True)

　　输出

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159

Model: "model" __________________________________________________________________________________________________ Layer (type)                    Output Shape         Param #     Connected to                     ================================================================================================== input_1 (InputLayer)            [(None, 160, 160, 3) 0                                            __________________________________________________________________________________________________ conv2d (Conv2D)                 (None, 80, 80, 32)   896         input_1[0][0]                    __________________________________________________________________________________________________ batch_normalization (BatchNorma (None, 80, 80, 32)   128         conv2d[0][0]                     __________________________________________________________________________________________________ activation (Activation)         (None, 80, 80, 32)   0           batch_normalization[0][0]        __________________________________________________________________________________________________ activation_1 (Activation)       (None, 80, 80, 32)   0           activation[0][0]                 __________________________________________________________________________________________________ separable_conv2d (SeparableConv (None, 80, 80, 64)   2400        activation_1[0][0]               __________________________________________________________________________________________________ batch_normalization_1 (BatchNor (None, 80, 80, 64)   256         separable_conv2d[0][0]           __________________________________________________________________________________________________ activation_2 (Activation)       (None, 80, 80, 64)   0           batch_normalization_1[0][0]      __________________________________________________________________________________________________ separable_conv2d_1 (SeparableCo (None, 80, 80, 64)   4736        activation_2[0][0]               __________________________________________________________________________________________________ batch_normalization_2 (BatchNor (None, 80, 80, 64)   256         separable_conv2d_1[0][0]         __________________________________________________________________________________________________ max_pooling2d (MaxPooling2D)    (None, 40, 40, 64)   0           batch_normalization_2[0][0]      __________________________________________________________________________________________________ conv2d_1 (Conv2D)               (None, 40, 40, 64)   2112        activation[0][0]                 __________________________________________________________________________________________________ add (Add)                       (None, 40, 40, 64)   0           max_pooling2d[0][0]                                                                               conv2d_1[0][0]                   __________________________________________________________________________________________________ activation_3 (Activation)       (None, 40, 40, 64)   0           add[0][0]                        __________________________________________________________________________________________________ separable_conv2d_2 (SeparableCo (None, 40, 40, 128)  8896        activation_3[0][0]               __________________________________________________________________________________________________ batch_normalization_3 (BatchNor (None, 40, 40, 128)  512         separable_conv2d_2[0][0]         __________________________________________________________________________________________________ activation_4 (Activation)       (None, 40, 40, 128)  0           batch_normalization_3[0][0]      __________________________________________________________________________________________________ separable_conv2d_3 (SeparableCo (None, 40, 40, 128)  17664       activation_4[0][0]               __________________________________________________________________________________________________ batch_normalization_4 (BatchNor (None, 40, 40, 128)  512         separable_conv2d_3[0][0]         __________________________________________________________________________________________________ max_pooling2d_1 (MaxPooling2D)  (None, 20, 20, 128)  0           batch_normalization_4[0][0]      __________________________________________________________________________________________________ conv2d_2 (Conv2D)               (None, 20, 20, 128)  8320        add[0][0]                        __________________________________________________________________________________________________ add_1 (Add)                     (None, 20, 20, 128)  0           max_pooling2d_1[0][0]                                                                             conv2d_2[0][0]                   __________________________________________________________________________________________________ activation_5 (Activation)       (None, 20, 20, 128)  0           add_1[0][0]                      __________________________________________________________________________________________________ separable_conv2d_4 (SeparableCo (None, 20, 20, 256)  34176       activation_5[0][0]               __________________________________________________________________________________________________ batch_normalization_5 (BatchNor (None, 20, 20, 256)  1024        separable_conv2d_4[0][0]         __________________________________________________________________________________________________ activation_6 (Activation)       (None, 20, 20, 256)  0           batch_normalization_5[0][0]      __________________________________________________________________________________________________ separable_conv2d_5 (SeparableCo (None, 20, 20, 256)  68096       activation_6[0][0]               __________________________________________________________________________________________________ batch_normalization_6 (BatchNor (None, 20, 20, 256)  1024        separable_conv2d_5[0][0]         __________________________________________________________________________________________________ max_pooling2d_2 (MaxPooling2D)  (None, 10, 10, 256)  0           batch_normalization_6[0][0]      __________________________________________________________________________________________________ conv2d_3 (Conv2D)               (None, 10, 10, 256)  33024       add_1[0][0]                      __________________________________________________________________________________________________ add_2 (Add)                     (None, 10, 10, 256)  0           max_pooling2d_2[0][0]                                                                             conv2d_3[0][0]                   __________________________________________________________________________________________________ activation_7 (Activation)       (None, 10, 10, 256)  0           add_2[0][0]                      __________________________________________________________________________________________________ conv2d_transpose (Conv2DTranspo (None, 10, 10, 256)  590080      activation_7[0][0]               __________________________________________________________________________________________________ batch_normalization_7 (BatchNor (None, 10, 10, 256)  1024        conv2d_transpose[0][0]           __________________________________________________________________________________________________ activation_8 (Activation)       (None, 10, 10, 256)  0           batch_normalization_7[0][0]      __________________________________________________________________________________________________ conv2d_transpose_1 (Conv2DTrans (None, 10, 10, 256)  590080      activation_8[0][0]               __________________________________________________________________________________________________ batch_normalization_8 (BatchNor (None, 10, 10, 256)  1024        conv2d_transpose_1[0][0]         __________________________________________________________________________________________________ up_sampling2d_1 (UpSampling2D)  (None, 20, 20, 256)  0           add_2[0][0]                      __________________________________________________________________________________________________ up_sampling2d (UpSampling2D)    (None, 20, 20, 256)  0           batch_normalization_8[0][0]      __________________________________________________________________________________________________ conv2d_4 (Conv2D)               (None, 20, 20, 256)  65792       up_sampling2d_1[0][0]            __________________________________________________________________________________________________ add_3 (Add)                     (None, 20, 20, 256)  0           up_sampling2d[0][0]                                                                               conv2d_4[0][0]                   __________________________________________________________________________________________________ activation_9 (Activation)       (None, 20, 20, 256)  0           add_3[0][0]                      __________________________________________________________________________________________________ conv2d_transpose_2 (Conv2DTrans (None, 20, 20, 128)  295040      activation_9[0][0]               __________________________________________________________________________________________________ batch_normalization_9 (BatchNor (None, 20, 20, 128)  512         conv2d_transpose_2[0][0]         __________________________________________________________________________________________________ activation_10 (Activation)      (None, 20, 20, 128)  0           batch_normalization_9[0][0]      __________________________________________________________________________________________________ conv2d_transpose_3 (Conv2DTrans (None, 20, 20, 128)  147584      activation_10[0][0]              __________________________________________________________________________________________________ batch_normalization_10 (BatchNo (None, 20, 20, 128)  512         conv2d_transpose_3[0][0]         __________________________________________________________________________________________________ up_sampling2d_3 (UpSampling2D)  (None, 40, 40, 256)  0           add_3[0][0]                      __________________________________________________________________________________________________ up_sampling2d_2 (UpSampling2D)  (None, 40, 40, 128)  0           batch_normalization_10[0][0]     __________________________________________________________________________________________________ conv2d_5 (Conv2D)               (None, 40, 40, 128)  32896       up_sampling2d_3[0][0]            __________________________________________________________________________________________________ add_4 (Add)                     (None, 40, 40, 128)  0           up_sampling2d_2[0][0]                                                                             conv2d_5[0][0]                   __________________________________________________________________________________________________ activation_11 (Activation)      (None, 40, 40, 128)  0           add_4[0][0]                      __________________________________________________________________________________________________ conv2d_transpose_4 (Conv2DTrans (None, 40, 40, 64)   73792       activation_11[0][0]              __________________________________________________________________________________________________ batch_normalization_11 (BatchNo (None, 40, 40, 64)   256         conv2d_transpose_4[0][0]         __________________________________________________________________________________________________ activation_12 (Activation)      (None, 40, 40, 64)   0           batch_normalization_11[0][0]     __________________________________________________________________________________________________ conv2d_transpose_5 (Conv2DTrans (None, 40, 40, 64)   36928       activation_12[0][0]              __________________________________________________________________________________________________ batch_normalization_12 (BatchNo (None, 40, 40, 64)   256         conv2d_transpose_5[0][0]         __________________________________________________________________________________________________ up_sampling2d_5 (UpSampling2D)  (None, 80, 80, 128)  0           add_4[0][0]                      __________________________________________________________________________________________________ up_sampling2d_4 (UpSampling2D)  (None, 80, 80, 64)   0           batch_normalization_12[0][0]     __________________________________________________________________________________________________ conv2d_6 (Conv2D)               (None, 80, 80, 64)   8256        up_sampling2d_5[0][0]            __________________________________________________________________________________________________ add_5 (Add)                     (None, 80, 80, 64)   0           up_sampling2d_4[0][0]                                                                             conv2d_6[0][0]                   __________________________________________________________________________________________________ activation_13 (Activation)      (None, 80, 80, 64)   0           add_5[0][0]                      __________________________________________________________________________________________________ conv2d_transpose_6 (Conv2DTrans (None, 80, 80, 32)   18464       activation_13[0][0]              __________________________________________________________________________________________________ batch_normalization_13 (BatchNo (None, 80, 80, 32)   128         conv2d_transpose_6[0][0]         __________________________________________________________________________________________________ activation_14 (Activation)      (None, 80, 80, 32)   0           batch_normalization_13[0][0]     __________________________________________________________________________________________________ conv2d_transpose_7 (Conv2DTrans (None, 80, 80, 32)   9248        activation_14[0][0]              __________________________________________________________________________________________________ batch_normalization_14 (BatchNo (None, 80, 80, 32)   128         conv2d_transpose_7[0][0]         __________________________________________________________________________________________________ up_sampling2d_7 (UpSampling2D)  (None, 160, 160, 64) 0           add_5[0][0]                      __________________________________________________________________________________________________ up_sampling2d_6 (UpSampling2D)  (None, 160, 160, 32) 0           batch_normalization_14[0][0]     __________________________________________________________________________________________________ conv2d_7 (Conv2D)               (None, 160, 160, 32) 2080        up_sampling2d_7[0][0]            __________________________________________________________________________________________________ add_6 (Add)                     (None, 160, 160, 32) 0           up_sampling2d_6[0][0]                                                                             conv2d_7[0][0]                   __________________________________________________________________________________________________ conv2d_8 (Conv2D)               (None, 160, 160, 3)  867         add_6[0][0]                      ================================================================================================== Total params: 2,058,979 Trainable params: 2,055,203 Non-trainable params: 3,776 __________________________________________________________________________________________________

　　结构图

　　netron.app 示例 

本地版安装教程：https://github.com/lutzroeder/netron