Data Augmentation

Learning how to effectively code a neural network is one of the steps involved in developing well-performing solutions. Additionally, to develop great deep learning solutions, it is crucial to find an area of interest in which we can provide a solution to a current challenge. But once all of that is done, we are typically faced with the same issue: getting a dataset of a decent size to get good performance from our models, either by self-gathering or by downloading it from the internet and other available sources.

As you might imagine, and even though it is now possible to gather and store vast amounts of data, this is not an easy task due to the costs associated with it. And so, most of the time, we are stuck working with a dataset containing tens of thousands of entries, and even fewer when referring to images.

This becomes a relevant issue when developing a solution for a computer vision problem, mainly due to two reasons:

• The larger the dataset...