Introduction

In the previous chapter, we looked at a traditional deep feedforward neural network. One of the limitations of a traditional deep feedforward neural network is that it is not translation-invariant, that is, a cat image in the upper-right corner of an image would be considered different from an image that has a cat in the center of the image. Additionally, traditional neural networks are affected by the scale of an object. If the object is big in the majority of the images and a new image has the same object in it but with a smaller scale (occupies a smaller portion of the image), traditional neural networks are likely to fail in classifying the image.

Convolutional Neural Networks (CNNs) are used to deal with such issues. Given that a CNN is able to deal with translation in images and also the scale of images, it is considered a lot more useful in object classification/ detection.

In this chapter, you will learn about the following:

• Inaccuracy of traditional neural network when images are translated
• Building a CNN from scratch using Python
• Using CNNs to improve image classification on a MNIST dataset
• Implementing data augmentation to improve network accuracy
• Gender classification using CNNs