如何在 keras ResNet 之后應用密集層？

Question

在 ResNet50 之后如何應用 Dense 層？ 這是我的代碼

    def build_model():
       x = tf.keras.applications.ResNet50(input_shape=(IMG_WIDTH, IMG_HEIGHT, 3), weights=None, 
       include_top=False, pooling='avg')
       model = tf.keras.layers.Dense(196)(x)
       model.summary()
       return model

但我收到了這個錯誤：

TypeError: Inputs to a layer should be tensors.

Answer 1

You can access the output of the model with the property output of the model, if you are willing to recreate a model using the functional API. 在這種情況下，使用 Sequential API 可能會更容易：

順序 API：

new_model = tf.keras.Sequential(
    [
        tf.keras.applications.ResNet50(input_shape=(IMG_WIDTH, IMG_HEIGHT, 3), weights=None, include_top=False, pooling='avg'),
        tf.keras.layers.Dense(196)
    ]
)
new_model.summary()

功能 API：

resnet = tf.keras.applications.ResNet50(input_shape=(IMG_WIDTH, IMG_HEIGHT, 3), weights=None, include_top=False, pooling='avg')
out = tf.keras.layers.Dense(196)(resnet.output)
new_model = tf.keras.models.Model(inputs=resnet.input, outputs=out)
new_model.summary()