[SOLVED] How to input last layer of pretrained network as tensor to flatten in Keras

How to input last layer of pretrained network as tensor to flatten in Keras

I'm trying to make a model with two outputs, one for classification and other for regression. I'm using pre-trained VGG16 from Keras as the convolutional feature extractor. But when I try to build the next flatten layer in both the output sequence, I get error that the input is not a tensor. How do I make sure I'm feeding only the last layer of the convolutional base as input to the next layers?

conv_base = VGG16(include_top = False,
                  weights = 'imagenet')
conv_base.trainable = False
with tf.variable_scope('classification_head', reuse=tf.AUTO_REUSE):
    x = Flatten()(conv_base.layers[-1])
    x = Dense()(x)

Error :

ValueError: Layer flatten_1 was called with an input that isn't a symbolic tensor. Received type: class 'keras.layers.pooling.MaxPooling2D'. Full input: [keras.layers.pooling.MaxPooling2D object at 0xc2855c9b0]. All inputs to the layer should be tensors.

Solution

Got the required tensor as :

features = conv_base.output 
print(features)

Tensor("block5_pool/MaxPool:0", shape=(?, 11, 14, 512), dtype=float32)