Keras: Removing layers from model gives the same output as original model

[koul]

Hello there,

During some feature extraction experiments, I noticed that the 'model.pop()' functionality is not working as expected. For a pretrained model like vgg16, after using 'model.pop()' , model.summary() shows that the layer has been removed (expected 4096 features), however on passing an image through the new model, it results in the same number of features (1000) as the original model. No matter how many layers are removed including a completely empty model, it generates the same output. See short example here, full code is attached at the end of this post. Looking for your guidance on what might be the issue.

#Passing an image through the full vgg16 model
model = VGG16(weights = 'imagenet', include_top = True, input_shape = (224,224,3))
img = image.load_img( 'cat.jpg', target_size=(224,224) )
img = image.img_to_array( img )
img = np.expand_dims( img, axis=0 )
img = preprocess_input( img )
features = model.predict( img )
features = features.flatten()
print(len(features)) #Expected 1000 features corresponding to 1000 imagenet classes

1000

model.layers.pop()
img = image.load_img( 'cat.jpg', target_size=(224,224) )
img = image.img_to_array( img )
img = np.expand_dims( img, axis=0 )
img = preprocess_input( img )
features2 = model.predict( img )
features2 = features2.flatten()
print(len(features2)) #Expected 4096 features, but still getting 1000. Why?
#No matter how many layers are removed, the output is still 1000

1000

Thank you!

See full code here:
bug-feature-extraction.pdf

[koul]

Found the solution from https://github.com/keras-team/keras/issues/2371

from keras.models import Model

model.layers.pop()
model2 = Model(model.input, model.layers[-1].output)
model2.summary()

model2 behaves as expected.

[famunir]

Either you pop the layer before converting the model to sequential as given here or the better way is to use
model._layers.pop()
instead of
model.layers.pop()

[ismael-elatifi]

As I posted here https://github.com/tensorflow/tensorflow/issues/22479#issuecomment-574301477, try using model.pop(). It worked for me with TF 2.0.0.
Example to replace the last layer :

model = tf.keras.models.load_model(path_model)
model.pop()  # to remove last layer
new_model = tf.keras.Sequential([
    model,
    tf.keras.layers.Dense(units=nb_classes, activation="softmax", name="new_layer_name")
])
new_model.build((None, height, width, 3))