迁移学习后TensorFlow ResNet50模型中GradCAM所需卷积层的访问方法

我是深度学习领域新手，在将GradCAM应用于自行重新训练的TensorFlow ResNet50模型时遇到问题。我构建模型的代码如下：

IMG_SHAPE = IMG_SIZE + (3,)
base_model = tf.keras.applications.ResNet50(input_shape=IMG_SHAPE,
                                               include_top=False,
                                               weights='imagenet')

global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
feature_batch_average = global_average_layer(feature_batch)

prediction_layer = tf.keras.layers.Dense(5, activation='softmax')
prediction_batch = prediction_layer(feature_batch_average)

inputs = tf.keras.Input(shape=(224, 224, 3))
x = data_augmentation(inputs)
x = preprocess_input(x)
x = base_model(x, training=False)
x = global_average_layer(x)
x = tf.keras.layers.Dropout(0.2)(x)
outputs = prediction_layer(x)
model = tf.keras.Model(inputs, outputs)

base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=base_learning_rate),
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),
              metrics=['accuracy'])

模型摘要：

Model: "model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_2 (InputLayer)        [(None, 224, 224, 3)]     0         
                                                                 
 sequential (Sequential)     (None, 224, 224, 3)       0         
                                                                 
 tf.__operators__.getitem (S  (None, 224, 224, 3)      0         
 licingOpLambda)                                                 
                                                                 
 tf.nn.bias_add (TFOpLambda)  (None, 224, 224, 3)      0         
                                                                 
 resnet50 (Functional)       (None, 7, 7, 2048)        23587712  
                                                                 
 global_average_pooling2d (G  (None, 2048)             0         
 lobalAveragePooling2D)                                           
                                                                 
 dropout (Dropout)           (None, 2048)              0         
                                                                 
 dense (Dense)               (None, 5)                 10245     
                                                                 
=================================================================
Total params: 23,597,957
Trainable params: 19,463,173
Non-trainable params: 4,134,784
_________________________________________________________________

按照GradCAM教程要求，需获取最后一层卷积层输出，尝试用以下代码：

last_conv_layer = model.get_layer("conv5_block3_out")
last_conv_layer_model = tf.keras.Model(model.inputs, last_conv_layer.output)

但该卷积层嵌套在模型的"resnet50 (Functional)"子模型中，无法直接访问。仅使用未修改的ResNet50基础模型时才能访问该层，但不符合项目需求。

解决方法 #

因为conv5_block3_out层是嵌套在resnet50子模型内部的，你需要先获取这个子模型，再从中提取目标层：

1. 从整体模型中取出嵌套的ResNet50子模型：

base_model = model.get_layer('resnet50')

2. 从子模型中获取conv5_block3_out层：

last_conv_layer = base_model.get_layer('conv5_block3_out')

3. 构建GradCAM所需的输出模型：

last_conv_layer_model = tf.keras.Model(model.inputs, last_conv_layer.output)

或者你可以直接用一行代码完成层的获取：

last_conv_layer = model.get_layer('resnet50').get_layer('conv5_block3_out')

这样就能成功访问到嵌套的卷积层，继续完成GradCAM的后续步骤。

内容的提问来源于stack exchange，提问作者Skruff