Louvre/Monet transfer after 20K iterations

I submitted the Art_Generation_with_Neural_Style_Transfer assignment with 100% grade and the code cells in my notebook get the “All tests passed” result after I run them.

However, when I run the Louvre/Monet test case with 20K iterations and 0.001 learning rate, my output image is not at all similar to what is documented in the notebook. This is what I got:

In the post Neural Style transfer (Louvre / Monet) parameters?, user “ernie” got a much better output with the same notebook:


According to the notebook documentation, this expected output after 20K iterations:

Ernie’s outptut is closer to what the notebook expects, compared to what I get. How do I go about debugging this?


i noticed quite early looking at the pictures they did not look like a smooth evolution. then i looked at those methods and i fiddled with the methods a_G a_S etc…

and ran it again and the pictures looked a bit more sensible?

Which methods did you fix?

I tired this out in compute_layer_style_cost:

a_S = tf.reshape(tf.transpose(a_S, perm=[3, 1, 2, 0]), shape=[n_C, n_H * n_W])

But this did not make any difference to the output image.

other than posting the entire result i have no idea what i did :confused:

I sent you a private message, please take a look and let me know what you think.