Get keras middle layer output in sequential

Question

The code is like this. Variable 'bneck' is keras sequential. I want to get the output of middle layer.

...
x = bneck(x)
x = CBNModule(960, 1, 1, activation=HSwish())(x)  # 7 * 960
s32 = CBNModule(320, 1, 1, activation=HSwish())(x)  # 7 * 960 -> 7 * 320
s32 = CBNModule(24, 1, 1, activation=HSwish())(s32)  # 7 * 320 -> 7 * 24
s16 = k.layers.Add()([
    CBNModule(24, 1, 1, activation=HSwish())(bneck.layers[12].output),
    UpModule(24, 2)(s32)
])  # (14 * 160 -> 14 * 24) + (7 * 24 -> 14 * 24)
...
return keras.Model(inputs=[...], outputs=[...])

When i run model.summary(), I got error like this ValueError: Graph disconnected: cannot obtain value for tensor KerasTensor

The error occurs on line 6 bneck.layers[12].output. But when i replace line 1 x = bneck(x) with code

for layer in bneck.layers:
    x = layers(x)

there is no error. Why is that? What's the difference bettwen them.

You want to join sequential model output to functional API model, am I right? — Innat
– Innat, Commented Mar 18, 2021 at 2:54
Yes. Apart from the final output, i also want to get the output of middle layers in sequential. — zixin wang
– zixin wang, Commented Mar 18, 2021 at 7:13

Innat · Accepted Answer · 2021-03-18 10:20:34Z

At first, you have to create a feature extractor based on your desire output layer. Your graph gets disconnected in here bneck.layers[12].output. Let's say you have model A and model B. And you want some layer's output (let's say 2 layers) from model A and use them in model B to complete its architecture. To do that, you first create 2 feature extractor from model A as follows

extractor_one = Model(modelA.input, expected_layer_1.output)
extractor_two = Model(modelA.input, expected_layer_2.output)

Here I will walk you through a simple code example. There can be a more flexible and smart approach to do this but here is one of them. I will build a sequential model and train it on CIFAR10 and next, I will try to build a functional model where I will utilize some of the sequential model layers (just 2 of them) and train the complete model on CIFAR100.

import tensorflow as tf 

seq_model = tf.keras.Sequential(
    [
        tf.keras.Input(shape=(32, 32, 3)),
        tf.keras.layers.Conv2D(16, 3, activation="relu"),
        tf.keras.layers.Conv2D(32, 3, activation="relu"),
        tf.keras.layers.Conv2D(64, 3, activation="relu"),
        tf.keras.layers.Conv2D(128, 3, activation="relu"),
        tf.keras.layers.Conv2D(256, 3, activation="relu"),
        tf.keras.layers.GlobalAveragePooling2D(), 
        tf.keras.layers.Dense(10, activation='softmax')
     
    ]
)

seq_model.summary()

Trian on CIFAR10 data set

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()

# train set / data 
x_train = x_train.astype('float32') / 255
y_train = tf.keras.utils.to_categorical(y_train , num_classes=10)

print(x_train.shape, y_train.shape)

seq_model.compile(
          loss      = tf.keras.losses.CategoricalCrossentropy(),
          metrics   = tf.keras.metrics.CategoricalAccuracy(),
          optimizer = tf.keras.optimizers.Adam())
# fit 
seq_model.fit(x_train, y_train, batch_size=128, epochs=5, verbose = 2)

# -------------------------------------------------------------------
(50000, 32, 32, 3) (50000, 10)
Epoch 1/5
27s 66ms/step - loss: 1.2229 - categorical_accuracy: 0.5647
Epoch 2/5
26s 67ms/step - loss: 1.1389 - categorical_accuracy: 0.5950
Epoch 3/5
26s 67ms/step - loss: 1.0890 - categorical_accuracy: 0.6127
Epoch 4/5
26s 67ms/step - loss: 1.0475 - categorical_accuracy: 0.6272
Epoch 5/5
26s 67ms/step - loss: 1.0176 - categorical_accuracy: 0.6409

Now, let' say we want some output from this sequential model, let's say of the following two layers.

tf.keras.layers.Conv2D(64, 3, activation="relu") # (None, 26, 26, 64)   
tf.keras.layers.Conv2D(256, 3, activation="relu") # (None, 22, 22, 256)

To get them we first create two feature extractor from the sequential model

last_layer_outputs = tf.keras.Model(seq_model.input, seq_model.layers[-3].output)
last_layer_outputs.summary() # (None, 22, 22, 256)  

mid_layer_outputs = tf.keras.Model(seq_model.input, seq_model.layers[2].output)
mid_layer_outputs.summary() # (None, 26, 26, 64)

Optionally, if we want to freeze them we can do that too now. Freezing because we choose the same type of data set here. (CIFAR 10-100).

print('last layer output')
# just freezing first 2 layer 
for layer in last_layer_outputs.layers[:2]:
  layer.trainable = False

# checking 
for l in last_layer_outputs.layers:
    print(l.name, l.trainable)


print('\nmid layer output')
# freeze all layers
mid_layer_outputs.trainable = False

# checking 
for l in mid_layer_outputs.layers:
    print(l.name, l.trainable)

last layer output
input_11 False
conv2d_81 False
conv2d_82 False
conv2d_83 False
conv2d_84 True
conv2d_85 True

mid layer output
input_11 False
conv2d_81 False
conv2d_82 False
conv2d_83 False

Now, let's create a new model with functional API and use the above two feature extractors.

encoder_input = tf.keras.Input(shape=(32, 32, 3), name="img")
x = tf.keras.layers.Conv2D(16, 3, activation="relu")(encoder_input)

last_x = last_layer_outputs(encoder_input)
print(last_x.shape) # (None, 22, 22, 256)

mid_x = mid_layer_outputs(encoder_input)
mid_x = tf.keras.layers.Conv2D(32, kernel_size=3, strides=1)(mid_x)
print(mid_x.shape) # (None, 24, 24, 32)

last_x = tf.keras.layers.GlobalMaxPooling2D()(last_x)
mid_x = tf.keras.layers.GlobalMaxPooling2D()(mid_x)
print(last_x.shape, mid_x.shape) # (None, 256) (None, 32)

encoder_output = tf.keras.layers.Concatenate()([last_x, mid_x])
print(encoder_output.shape) # (None, 288)

encoder_output = tf.keras.layers.Dense(100, activation='softmax')(encoder_output)
print(encoder_output.shape) # (None, 100)

encoder = tf.keras.Model(encoder_input, encoder_output, name="encoder")
encoder.summary()

Train on CIFAR100

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar100.load_data()

# train set / data 
x_train = x_train.astype('float32') / 255
y_train = tf.keras.utils.to_categorical(y_train , num_classes=100)

print(x_train.shape, y_train.shape)

encoder.compile(
          loss      = tf.keras.losses.CategoricalCrossentropy(),
          metrics   = tf.keras.metrics.CategoricalAccuracy(),
          optimizer = tf.keras.optimizers.Adam())
# fit 
encoder.fit(x_train, y_train, batch_size=128, epochs=5, verbose = 1)

Reference: Feature extraction with a Sequential model

I have some new questions. In this case, does it mean that the two extractors run separately? Actually, for the same input image, I hope to run the model once to extract different features.
If you don't understand in any part, please feel free to ask.
I gave another answer based on your comment, if it what you want, please mark that as the right answer so that future reader can get understanding.

Innat · Accepted Answer · 2021-03-22 20:35:45Z

Based on your first comment on my first post, I'm adding a new post rather than editing my existing answer as it's already got a too-long post. Anyway, your concern is reasonable. Even I was also struggling with some kind of issue with subclasses API, here. But It seems that I didn't write quite well in my query there, as people didn't feel like it's a matter of concern.

Anyway, here is another with a more concise and precise answer as we build a single model with desired outputs. A single extractor rather than previously two separate extractors which brought extra computation overhead. Let's say, our sequential model

import tensorflow as tf 

seq_model = tf.keras.Sequential(
    [
        tf.keras.Input(shape=(32, 32, 3)),
        tf.keras.layers.Conv2D(16, 3, activation="relu", name='conv1'),
        tf.keras.layers.Conv2D(32, 3, activation="relu", name='conv2'),
        tf.keras.layers.Conv2D(64, 3, activation="relu", name='conv3'),
        tf.keras.layers.Conv2D(128, 3, activation="relu", name='conv4'),
        tf.keras.layers.Conv2D(256, 3, activation="relu", name='conv5'),
        tf.keras.layers.GlobalAveragePooling2D(), 
        tf.keras.layers.Dense(10, activation='softmax')
     
    ]
)

for l in seq_model.layers:
    print(l.name, l.output_shape)

conv1 (None, 30, 30, 16)
conv2 (None, 28, 28, 32)
conv3 (None, 26, 26, 64)
conv4 (None, 24, 24, 128)
conv5 (None, 22, 22, 256)
global_average_pooling2d_3 (None, 256)
dense_3 (None, 10)

And we want conv3 and conv5 from a single model. We can do that easily as follows

model = tf.keras.models.Model(
        inputs=[seq_model.input], 
        outputs=[seq_model.get_layer('conv3').output,
                 seq_model.get_layer('conv5').output]
    )

# check 
for i in check_model(tf.keras.Input((32, 32, 3))):
    print(i.name, i.shape)

model_13/conv3/Relu:0 (None, 26, 26, 64)
model_13/conv5/Relu:0 (None, 22, 22, 256)

Nice, two feature output from the expected layer. Now, let's use this two layers (like my first post) to build a functional API model.

encoder_input = tf.keras.Input(shape=(32, 32, 3), name="img")
x = tf.keras.layers.Conv2D(16, 3, activation="relu")(encoder_input)

last_x = check_model(encoder_input)[0]
print(last_x.shape) # (None, 26, 26, 64) - model_13/conv3/Relu:0 (None, 26, 26, 64)

mid_x = check_model(encoder_input)[1] # model_13/conv5/Relu:0 (None, 22, 22, 256)
mid_x = tf.keras.layers.Conv2D(32, kernel_size=3, strides=1)(mid_x)
print(mid_x.shape) # (None, 20, 20, 32)

last_x = tf.keras.layers.GlobalMaxPooling2D()(last_x)
mid_x = tf.keras.layers.GlobalMaxPooling2D()(mid_x)
print(last_x.shape, mid_x.shape) # (None, 64) (None, 32)

encoder_output = tf.keras.layers.Concatenate()([last_x, mid_x])
print(encoder_output.shape) # (None, 96)

encoder_output = tf.keras.layers.Dense(100, activation='softmax')(encoder_output)
print(encoder_output.shape) # (None, 100)

encoder = tf.keras.Model(encoder_input, encoder_output, name="encoder")

tf.keras.utils.plot_model(
    encoder,
    show_shapes=True,
    show_layer_names=True
)

(None, 26, 26, 64)
(None, 20, 20, 32)
(None, 64) (None, 32)
(None, 96)
(None, 100)

Collectives™ on Stack Overflow

Get keras middle layer output in sequential

2 Answers 2

3 Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related