Cannot run imagenet_pretrained_vgg example

[RXZ2020]

I was trying to run https://github.com/google/objax/blob/master/examples/image_classification/imagenet_pretrained_vgg.py

But received an error:

---

FileNotFoundError Traceback (most recent call last)
in ()
38
39 # Load model with pretrained weights and make a prediction.
---> 40 model = vgg.VGG19(pretrained=True)
41 logit = model(img)
42 prob = objax.functional.softmax(logit)[0]

/usr/local/lib/python3.6/dist-packages/objax/zoo/vgg.py in init(self, pretrained)
55 if not os.path.exists(_VGG19_NPY):
56 raise FileNotFoundError(
---> 57 'You must download vgg19.npy from %s and save it to %s' % (_VGG19_URL, _VGG19_NPY))
58 if not os.path.exists(_SYNSET_PATH):
59 request.urlretrieve(_SYNSET_URL, _SYNSET_PATH)

FileNotFoundError: You must download vgg19.npy from https://github.com/machrisaa/tensorflow-vgg and save it to ./objax/zoo/pretrained/vgg19.npy

Is this normal? How should I proceed from here.

[aterzis-google]

@RXZ2020 please download vgg19.npy from the URL mentioned in the error message and save it to ./objax/zoo/pretrained/vgg19.npy

[david-berthelot]

You know I've come to realize that's not really practical. If you install Objax with pip install, the path where to save may not be obvious.

I've seen other frameworks create their own area say $HOME/.objax, maybe it something we should think about.

[RXZ2020]

@RXZ2020 please download vgg19.npy from the URL mentioned in the error message and save it to ./objax/zoo/pretrained/vgg19.npy

Hi, how do you do this if I was running it on Colab? Do I create an objax folder in Google drive or elsewhere?

[david-berthelot]

Yes, good point. We should make Objax download it for you otherwise it won't be useful in a Colab.

[aterzis-google]

If installing Objax with pip install, then need to copy under site-packages/objax/zoo/pretrained.

Not sure what the copyright for pretrained VGG is and whether we can include it in the objax package.