12.1 Activation Optimization --imresize and pilatus800.jpg issues and solutions

[mikechen66]

I found out the two issues that stopped the example application running. I have corrected them as follows.

1. imresize issue:

Issue: ImportError: cannot import name 'imresize' from 'scipy.misc'

Solution:

First install scikit-image $ pip install scikit-image or $ conda install -c conda-forge scikit-image

And then change from scipy.misc import imresize to from skimage.transform import resize

Please have a look at the skimage example at the official website ad follows.

skimage: https://scikit-image.org/docs/stable/auto_examples/transform/plot_rescale.html

2. pilatus800.jpg issue

Issue: No directory of data/pilatus800.jpg

Solution:

Need to download pilatus800.jpg at the resource as follows. pilatus800.jpg: //github.com/Spidy20/Data-scince-ML-project/tree/master/Deep%20dream

And then save pilatus800.jpg in the directory of /dl-cookbook/data/.

At last change data/pilatus800.jpg to /home/user/Documents/dl-cookbook/data/pilatus800.jpg in the page of 12.1 Activation Optimization by jupyter notebook.

[mikechen66]

With regard to the RuntimeError, the small modification will make the example script run correctly.

RuntimeError: Variable += value not supported. Use variable.assign_add(value) to modify the variable value and variable = variable + value to get a new Tensor object.

1. Issue

---

RuntimeError Traceback (most recent call last) in 14 loss += coeff * K.sum(K.square(x[:, :, 2: -2, 2: -2])) / scaling 15 else: ---> 16 loss += coeff * K.sum(K.square(x[:, 2: -2, 2: -2, :])) / scaling 17 18 grads = K.gradients(loss, input_img)[0]

~/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/ops/resource_variable_ops.py in iadd(self, unused_other) 1220 1221 def iadd(self, unused_other): -> 1222 raise RuntimeError("Variable += value not supported. Use " 1223 "variable.assign_add(value) to modify the variable " 1224 "value and variable = variable + value to get a new "

RuntimeError: Variable += value not supported. Use variable.assign_add(value) to modify the variable value and variable = variable + value to get a new Tensor object.

2. Solution

Change the original scripts

    if K.image_data_format() == 'channels_first':
        loss += coeff * K.sum(K.square(x[:, :, 2: -2, 2: -2])) / scaling
    else:
        loss += coeff * K.sum(K.square(x[:, 2: -2, 2: -2, :])) / scaling

to the following scripts.

    if K.image_data_format() == 'channels_first':
        loss = loss + coeff * K.sum(K.square(x[:, :, 2: -2, 2: -2])) / scaling
    else:
        loss = loss + coeff * K.sum(K.square(x[:, 2: -2, 2: -2, :])) / scaling