CODEWITHSUNDEEP
pythontensorflowdeep-learningneural-networkclassification
abcdefghi999955
abcdefghi999955

Reputation: 21

Get softmax output and raw output of the last layer of a model

When creating a neural network for image classification, I want to get the classification on one hand and the raw output on the other hand to determine if the image really contains one of the images I want to classify or not. If not then the raw output should contain very low values for all classes. But if the image really contains one of the objects that I want to classify, then the raw output should have a high value for one of the neurons.

Assuming I have the following code:

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(16, (3, 3), activation='relu', input_shape=(80, 80, 3)))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Dropout(0.3))
model.add(tf.keras.layers.Conv2D(16, (3, 3), activation='relu'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Dropout(0.3))
model.add(tf.keras.layers.Conv2D(16, (3, 3), activation='relu'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(4, activation='softmax'))

How would I get the raw output of the last dense layer?

Upvotes: 0

Views: 567

Answers (1)

user3797387
user3797387

Reputation:

You can use functional API and implement your model in a next way:

  inputs = tf.keras.Input(shape=(80, 80, 3))
  x = tf.keras.layers.Conv2D(16, (3, 3), activation='relu')(inputs)
  x = tf.keras.layers.MaxPooling2D((2, 2))(x)
  x = tf.keras.layers.Dropout(0.3)(x)
  x = tf.keras.layers.Conv2D(16, (3, 3), activation='relu')(x)
  x = tf.keras.layers.MaxPooling2D((2, 2))(x)
  x = tf.keras.layers.Dropout(0.3)(x)
  x = tf.keras.layers.Conv2D(16, (3, 3), activation='relu')(x)
  x = tf.keras.layers.MaxPooling2D((2, 2))(x)
  x = tf.keras.layers.Flatten()(x)
  # here you can get raw output
  logits = tf.keras.layers.Dense(4)(x)
  model = tf.keras.Model(
    inputs=inputs, 
    outputs={
      'logits': logits,
      'predictions': tf.nn.softmax(logits)
    }
  )
  model.summary()

After that, your model will have two outputs in dictionary format. Beware that you can't use a simple loss function like categorical_crossentropy because it will try to minimize loss for both outputs. You need to use losses argument in compile method to specify the loss for each output. For example:

  model.compile(
    optimizer='adam',
    loss={
      # ignore logits loss
      'logits': lambda y_true, y_pred: 0.0,
      'predictions': tf.keras.losses.CategoricalCrossentropy()
    })

And your fit would look like this:

  model.fit(
    x_train,
    {
      'logits': y_train,
      'predictions': y_train
    },
    epochs=10
  )

Upvotes: 1

Related Questions