Data augmentation is a methodology employed in computer vision to significantly expand the size of a training dataset. It is particularly useful when dealing with relatively small datasets and aims to enhance the generalization and robustness of a machine learning model. This technique involves applying various geometric transformations, both affine and non-affine, such as rotation, horizontal or vertical flipping, zooming, translation, and shearing. Additionally, adjustments in brightness and contrast, noise injection, color modifications, elastic transformations, and cropping are commonly utilized as well.
"Data augmentation is a set of techniques that enhance the size and quality of machine learning training datasets so that better deep learning models can be trained with them. Data Augmentation artificially inflates datasets using label-preserving data transformations."