Data Augmentation

While training machine learning models, we data scientists often run into the problem of imbalanced classes and a lack of training data. This leads to sub-par models that perform poorly when deployed in real-life scenarios. One easy way to deal with these problems is data augmentation. There are multiple ways of performing data augmentation, such as rotating the image, shifting the object, cropping an image, shearing to distort the image, and zooming in to a part of the image, as well as more complex methods such as using Generative Adversarial Networks (GANs) to generate new images. GANs are simply two neural networks that are competing with each other. A generator network tries to make images that are similar to the already existing images, while a discriminator network tries to determine if the image was generated or was part of the original data. After the training is complete, the generator network is able to create images that are not a part of the original data but...