Neural Networks with Keras Cookbook

Getting ready

In this recipe, we will be working on the CIFAR-10 dataset, which contains images of objects of 10 different classes.

The strategy that we'll use is as follows:

• Download the CIFAR-10 dataset
• Preprocess the dataset
  • Scale the input values
  • One-hot-encode the output classes
• Build a deep CNN with multiple convolution and pooling layers
• Compile and fit the model to test its accuracy on the test dataset
• Generate random translations of the original set of images in the training dataset
• Fit the same model architecture that was built in the previous step on the total images (generated images, plus the original images)
• Check the accuracy of the model on the test dataset

We will be implementing data augmentation using the ImageDataGenerator ...