来自 keras 密集层的意外 output 形状

Question

I try to create a minimal non-convolutional NN image binary classifier with one hidden layer only (as a practice before more complicated models):

def make_model(input_shape):
    inputs = keras.Input(shape=input_shape)
    x = layers.Dense(128, activation="ReLU")(inputs)
    outputs = layers.Dense(1, activation="sigmoid")(x)
    return keras.Model(inputs, outputs)
model = make_model(input_shape=(256, 256, 3))

Its model.summary() shows

Model: "model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_1 (InputLayer)        [(None, 256, 256, 3)]     0                                                                       
 dense (Dense)               (None, 256, 256, 128)     512                                                                    
 dense_1 (Dense)             (None, 256, 256, 1)       129                                                      
=================================================================
Total params: 641
Trainable params: 641
Non-trainable params: 0

Since the dense_1 layer has one neuron only, what I expect from this layer is an output shape of (None, 1) (i,e, a single number indicating the predicted binary label) but instead the model gives (None, 256, 256, 1) .

What's wrong with my model setting and how can I get it right?

Answer 1

You have to flatten your preposterously large tensor if you want to use the output shape (None, 1) :

import tensorflow as tf

def make_model(input_shape):
    inputs = tf.keras.layers.Input(shape=input_shape)
    x = tf.keras.layers.Dense(128, activation="relu")(inputs)
    x = tf.keras.layers.Flatten()(x)
    outputs = tf.keras.layers.Dense(1, activation="sigmoid")(x)
    return tf.keras.Model(inputs, outputs)

model = make_model(input_shape=(256, 256, 3))
print(model.summary())

Answer 2

A mistake is in your function make_model .

def make_model(input_shape):
    inputs = keras.Input(shape=input_shape)
    x = layers.Dense(128, activation="ReLU")(x)
    outputs = layers.Dense(1, activation="sigmoid")(x)
    return keras.Model(inputs, outputs)

You probably wanted the second line to be

x = layers.Dense(128, activation="ReLU")(inputs)

and not

x = layers.Dense(128, activation="ReLU")(x)

and unfortunately, x exists in scope, so it didn't throw an error.