'attributeError: 'Tensor' object has no attribute '_keras_history' during implementing perceptual loss with pretrained VGG using keras

Question

I'm trying to implement the VGG perceptual loss for a model training for video inputs. I implemented the perceptual loss like the recommendation in the question AttributeError: 'Tensor' object has no attribute '_keras_history':

My mainModel looks like the following graph: Graph of mainModel

The input size is (bathsize, frame_num, row, col, channel). And I want to get the perceptual loss for the middle frame, that is, frame_num/2.

So, I implemented the following lossModel:

lossModel = VGG19(weights='imagenet')
lossModel = Model(inputs=lossModel.input,outputs=lossModel.get_layer('block3_conv4').output)
lossOut = lossModel(mainModel.output[:,frame_num/2])
fullModel = Model(mainModel.input,lossOut)

But I faced an error message in the line fullModel = Model(mainModel.input, lossOut):

attributeError: 'Tensor' object has no attribute '_keras_history'

BTW, I'm using keras version is '2.0.9'.

Could anyone help me with this?

THANKS a lot!!

Answer 1

This most of the times means that you're doing calculations outside layers.

A keras model needs to be made of keras layers. Operations outside layers are not allowed.

Take all your calculations and put them inside Lambda layers: https://keras.io/layers/core/#lambda

---

Here, the mainModel.output[:,frame_num/2] is an operation outside a layer.

Transfer it to a Lambda layer:

lossModel = VGG19(weights='imagenet')
lossModel = Model(inputs=lossModel.input,outputs=lossModel.get_layer('block3_conv4').output)

#you must connect lossmodel and mainmodel somewhere!!!
output = lossModel(mainModel.output)

output = Lambda(lambda x: x[:,frame_num/2])(output)

fullModel = Model(mainModel.input, output)