C4 W4 A2 - Neural Style Transfer

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I’m attempting to implement the tensor flow train_step() func. I keep getting the following error:
I’ve tried looking at the docs as well as the gradient tape lecture, but still can’t resolve. Please assist.

ValueError: in user code:

<ipython-input-133-effcf516998e>:34 train_step  *
    optimizer.apply_gradients([(grad, generated_image)])
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/optimizer_v2/optimizer_v2.py:513 apply_gradients  **
    grads_and_vars = _filter_grads(grads_and_vars)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/optimizer_v2/optimizer_v2.py:1271 _filter_grads
    ([v.name for _, v in grads_and_vars],))

ValueError: No gradients provided for any variable: ['Variable:0'].
1 Like
2

Please click my name and message your notebook as an attachment.

3

Gradient tape is quite useful tool, but you need to be very careful that target variables need to be in the scope of “with tf.GradientTape() as tape:”.
In this case, J seems to be a problem. During waiting for balaji’s inspection, please check an indent of J =… It needs to be the same level as J_style, J_content,… i.e, within the scope of GradientTape.

4

Thanks. Checked the indentation is good–J does get computed within the context of the gradient. The call to tape is outside the context though… but the hard-coded python doesn’t set the persistent flag, so I didn’t change any code outside of place your code here.

5

The call to tape is outside the context though…

That’s OK. So, please wait for balaji’s check.

6

There’s no need to do this: generated_image = tf.Variable(tf.image.convert_image_dtype(content_image, tf.float32)) when invoking vgg_model_outputs. Use the function parameter generated_image so that gradients are correctly tracked.

1 Like
7

It worked :laughing: Thanks Balaji