时间戳

首页 > 解决方案 > 在 Tensorflow 的 AlexNet 实现中查找输出节点

python - 在 Tensorflow 的 AlexNet 实现中查找输出节点

问题描述

我正在使用Pre-Trained Alexnet的这种实现。我已经用我自己的训练数据重新训练了它，并通过这样做我收到了重新训练的AlexNet的检查点。

我现在想用它来分类图像。据我了解，我需要为分类提供两件事：

  1. 带有我要分类的图像的输入节点
  2. 结果的输出节点

我收到所有节点的列表，使用

for op in graph.get_operations():
    print(str(op.name), op.outputs)

但是，我无法以这种方式找出输出节点。所有节点的列表可以在这里看到。

我的方法错了吗？我对tensorflow不是很有经验，我很感激任何帮助。非常感谢。

标签： pythontensorflowneural-networkconv-neural-network

解决方案

调用model.summary()以打印有用的模型摘要，其中包括：

  • 模型中所有层的名称和类型。
  • 每层的输出形状。
  • 每层的权重参数个数。
  • 每层接收的输入
  • 模型的可训练和不可训练参数的总数。

此外，您可以使用model.layers[]打印图层信息。

示例：我在这里定义了一个简单的模型，并model.summary()使用model.layers[].

import tensorflow as tf
from tensorflow.python.keras import Sequential
from tensorflow.keras.layers import MaxPooling2D, Conv2D, Dense, Dropout, Flatten, BatchNormalization
from tensorflow.keras.models import load_model


# Add the layers 
model = Sequential()
model.add(Conv2D(64,(3,3), input_shape=(424,424,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(3,3)))
model.add(Dense(32, activation='relu'))
model.add(Conv2D(64,(3,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(3,3)))
model.add(Dense(64, activation='relu'))
model.add(Dropout(.3))#test
model.add(Conv2D(64,(3,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(3,3)))
model.add(Dropout(.3))
model.add(Flatten(input_shape=(424,424,3)))
model.add(BatchNormalization())
model.add(Dense(2, activation='softmax'))

# Model summary 
model.summary()

输出 -

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_2 (Conv2D)            (None, 422, 422, 64)      1792      
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 140, 140, 64)      0         
_________________________________________________________________
dense_1 (Dense)              (None, 140, 140, 32)      2080      
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 138, 138, 64)      18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 46, 46, 64)        0         
_________________________________________________________________
dense_2 (Dense)              (None, 46, 46, 64)        4160      
_________________________________________________________________
dropout (Dropout)            (None, 46, 46, 64)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 44, 44, 64)        36928     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 14, 14, 64)        0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 14, 14, 64)        0         
_________________________________________________________________
flatten (Flatten)            (None, 12544)             0         
_________________________________________________________________
batch_normalization (BatchNo (None, 12544)             50176     
_________________________________________________________________
dense_3 (Dense)              (None, 2)                 25090     
=================================================================
Total params: 138,722
Trainable params: 113,634
Non-trainable params: 25,088
_________________________________________________________________

在构建模型后打印图层信息 -

# To print all the layers of the Model
print("All the Layers of the Model:")
for layers in model.layers:
    print(layers)
print("\n")

# To print first layer OR Input layer of the Model
print("Input Layer of the Model:","\n",model.layers[0],"\n")

# To print last layer OR Output layer of the Model
print("Output Layer of the Model:","\n",model.layers[-1])

输出 -

All the Layers of the Model:
<tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7faa09294550>
<tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7faa09294cf8>
<tensorflow.python.keras.layers.core.Dense object at 0x7faa09294d30>
<tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7faa0044e780>
<tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7faa0046fda0>
<tensorflow.python.keras.layers.core.Dense object at 0x7faa09294f98>
<tensorflow.python.keras.layers.core.Dropout object at 0x7faa00477da0>
<tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7faa0046ff60>
<tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7faa004412b0>
<tensorflow.python.keras.layers.core.Dropout object at 0x7faa00477f60>
<tensorflow.python.keras.layers.core.Flatten object at 0x7faa004802b0>
<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7faa003d0b00>
<tensorflow.python.keras.layers.core.Dense object at 0x7faa00441588>


Input Layer of the Model: 
 <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7faa09294550> 

Output Layer of the Model: 
 <tensorflow.python.keras.layers.core.Dense object at 0x7faa00441588>

推荐阅读