DLS Course 4 W2 A2 Input layer shape is not correct

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I am working on Exercise 2 the alpaca model and run into a problem with the input layer.

I get the following message:

Test failed

Expected value

[‘InputLayer’, [(None, 160, 160, 3)], 0]

does not match the input value:

[‘InputLayer’, [(1, 160, 160, 3)], 0]

With debugging I have found that the statement:

# create the input layer (Same as the imageNetv2 input size)

Returns a tensorflow object with the shape (None, 160, 160, 3)

The following statement:

# apply data augmentation to the inputs

Returns a tensorflow object with the shape (1, 160, 160, 3)

I do not know how to change the augmentation statement in order to have it return shape (None, 160, 160, 3)

Does anyone has a hint for me?

2

You code in alpaca_model() should call the data_augmentation() function.

3

I have programmed the model so that it calls the data_augmentation() function in the step:

# apply data augmentation to the inputs

x = data_augmentation…

the statement print(x) gives me: Tensor(“sequential_1/random_rotation_1/transform_28/ImageProjectiveTransformV2:0”, shape=(1, 160, 160, 3), dtype=float32)

4

The parameter passed to data_augmentation(…) should be “inputs”.

“inputs” should be a Keras Input object with the shape equal to the input_shape.

Tip: Don’t make things too complicated. There’s nothing tricky going on here.

5

Thanks for helping me out, but I have done that already. I have the following lines of code:

inputs = tf.keras.Input(shape=input_shape)
print(inputs)
x = data_augmentation(inputs)
print(x)

The print statements give:

Tensor("input_85:0", shape=(None, 160, 160, 3), dtype=float32)
Tensor("sequential_1/random_rotation_1/transform_38/ImageProjectiveTransformV2:0", shape=(1, 160, 160, 3), dtype=float32)

Because of the error message i described earlier my thought is that it is going wrong here.

6

Did you happen to modify this line of code?

image
image846×358 17.4 KB

If yes, that will cause problems.

8

Hi TMosh,

Thank you for your support. I did not change that line of code.

But now after a Kernel restart, the code runs without error. Not sure what the problem was, but I cannot recreate it. Thanks again for your help.

2 Likes
9

In a lot of these assignments in C4 and C5, the notebooks use global objects. So while you’re editing the notebook, any changes you make that modify a global object can cause runtime problems with previously tested code.

Often restarting the kernel and running all the cells again will straighten out these issues.