How to extract features from a pretrained model without calling `predict()` on any input?

Keras
,
1

I am migrating a legacy app from Torch which uses VGG19 pretrained model. The following code:

vgg_features = VGG19(weights='imagenet', include_top=False)
self.f1 = Sequential(*[vgg_features[x] for x in range(2)])

hits the following error:

    self.f1 = Sequential(*[vgg_features[x] for x in range(2)])
                           ~~~~~~~~~~~~^^^
TypeError: 'Functional' object is not subscriptable

The legacy PyTorch code that I am trying to migrate from is:

vgg_features = models.vgg19(pretrained=True).features

which doesn’t need to call predict() on any input data at all.