Large-scale visual recognition with GoogLeNet/Inception

 In 2014, the paper Going Deeper with Convolutions (https://arxiv.org/abs/1409.4842) was published by Google, introducing the architecture. Subsequently, newer versions (https://arxiv.org/abs/1512.00567 in 2015) were published under the name Inception. In these GoogLeNet/Inception models, multiple convolutional layers are applied in parallel before being stacked and fed to the layer. A great benefit of the network architecture is that the computational cost is lower and the file size of the trained weights is much smaller. In this recipe, we will demonstrate how to load the InceptionV3 weights in Keras and apply the model to classify images.

How to do it...

1. Keras has some great tools for using pretrained models. We start with importing the libraries and tools, as follows:

import numpy as np

from keras.applications.inception_v3 import InceptionV3
from keras.applications import imagenet_utils
from keras.preprocessing.image import load_img...