Week 4, assignment 2 train_step

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I did this

        a_G = vgg_model_outputs(generated_image)
        
        # Compute the style cost
        #(1 line)
        J_style = compute_style_cost(a_S,a_G)

        #(2 lines)
        # Compute the content cost
        J_content = compute_content_cost(a_C, a_G)
        # Compute the total cost
        J = total_cost(J_content, J_style, alpha = 10, beta = 40)

and am getting an error like this:

ValueError: Cannot reshape a tensor with 48 elements to shape [1,625,512] (320000 elements) for ā€˜{{node Reshape}} = Reshape[T=DT_FLOAT, Tshape=DT_INT32](strided_slice_1, Reshape/shape)ā€™ with input shapes: [4,4,3], [3] and with input tensors computed as partial shapes: input[1] = [1,625,512].

Blockquote

can someone help me out?

1 Like
2

The code you show looks the same as mine, other than that I passed STYLE_LAYERS to the compute_style_cost function. But when I do it the way you show, I donā€™t get the error messages that you show.

I can only guess that maybe something is wrong with one of your earlier functions. Are you sure you actually looked through the output of all the cells and didnā€™t just click ā€œValidateā€?

3

I would like to add something to this. I was in a similar situation than @paulinpaloalto and the assert line were wrong but my code was correct. Every time I ran the code, a different value of J appeared!. A simple but efective thing to try is ā€œRestart and Clear Outputā€ and just run again, everything worked.

4

Note that they tell you that executing the training is ā€œstatefulā€. Once youā€™ve run it, you have to reset the train_step cell before you can run it again.