ValueError: Layer weight shape (3, 3, 3, 64) not compatible with provided weight shape (64, 3, 3, 3)

[chikiuso]

Hi, I got the following message running the command, it seems to be the weight shape problem , any idea on solving it? thanks :

Traceback (most recent call last):
File "/usr/local/bin/make_image_analogy.py", line 27, in
image_analogy.main.main(args, model_class)
File "/usr/local/lib/python2.7/dist-packages/image_analogy/main.py", line 69, in main
net = vgg16.get_model(img_width, img_height, weights_path=args.vgg_weights, pool_mode=args.pool_mode)
File "/usr/local/lib/python2.7/dist-packages/image_analogy/vgg16.py", line 89, in get_model
layer.set_weights(weights)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1154, in set_weights
'provided weight shape ' + str(w.shape))
ValueError: Layer weight shape (3, 3, 3, 64) not compatible with provided weight shape (64, 3, 3, 3)

[awentzonline]

I suspect you have keras 2 installed, but this only runs on v1. I need to update the versions in setup.py. To confirm, run pip show keras.

[chikiuso]

Hi creator of project, yes, my keras version is 2.02

[AhmadBaracat]

Hi @awentzonline, are you planning on supporting Keras 2 anytime soon? Is there a workaround for the time being?

[pmlantos]

Hi, I had the same question as above. I don't have a way to access anything below Keras 2 as I'm using a shared computing cluster. Any suggestions?
Thanks!

[AhmadBaracat]

@pmlantos What I have done is to use virtualenv and instal Keras==1.1.1 on it. Not sure if that is feasible in your case.

[pmlantos]

Thanks, it may be ... I've set up a virtual environment with 1.2.1. Do I need to reinstall image-analogies in the virtual environment too? On Apr 29, 2017 3:26 AM, "Ahmed Baracat" <notifications@github.com> wrote: @pmlantos <https://github.com/pmlantos> What I have done is to use virtualenv and instal Keras==1.1.1 on it. Not sure if that is feasible in your case. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#39 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AajKY8OQ9Stizg_xb3WcUmOhiv7mjwOrks5r0uY-gaJpZM4M1bVy> .

[pmlantos]

Unfortunately I can't get this package to run within a virtual environment, though setting the environment up seems straightforward. Wonder if I can find a way to cheaply do it on AWS.

[AhmadBaracat]

That's what worked for me. Hope it helps.

1. Activate the virtualenv
2. Check the version of keras installed using pip show keras
3. If it is Keras version 2 then pip uninstall keras followed by pip install keras==1.1.1.
4. Uninstall image-analogies
5. Reinstall image-analogies

[pmlantos]

Thanks, unfortunately doesn't work for me. Some of the installation has to go into shared parts of the computing cluster that I don't have access to. Appreciate the help though.

…

On Sun, Apr 30, 2017 at 5:37 AM, Ahmed Baracat ***@***.***> wrote: That's what worked for me. Hope it helps. 1. Activate the virtualenv 2. Check the version of keras installed using pip show keras 3. If it is Keras version 2 then pip uninstall keras followed by pip install keras==1.1.1. 4. Uninstall image-analogies 5. Reinstall image-analogies — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#39 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AajKY7SbYca6ke1ZGs0AHy1Wbm_dkBhfks5r1FZegaJpZM4M1bVy> .

[AhmadBaracat]

No worries, but I think that if you are using virtualenv then you don't need to depend on shared parts because you get to create an isolated environment.

[oceanjack]

I have same problem as first

make sure that

requirements.txt

Cython==0.23.4
h5py==2.5.0
Keras==1.1.1
numpy==1.10.4
Pillow==3.1.1
PyYAML==3.11
scipy==0.17.0
scikit-learn==0.17.0
six==1.10.0
-e git://github.com/Theano/Theano.git@954c3816a40de172c28124017a25387f3bf551b2#egg=Theano

and I use sudo pip install theano==0.9.0b1

than it works~

[oceanjack]

and make sure that you have edit ~/.keras/keras.json like this
{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "theano"
}

[HeadstrongCatgirl]

I have replaced the 94th line in image_analogy/vgg16.py with
weights[0] = np.transpose(np.array(weights[0])[:, :, ::-1, ::-1], (2, 3, 1, 0))
and it works like charm now.