Data Augmentation Using GANs for 3D Applications

I. Maniadis
V. Solachidis
N. Vretos
P. Daras
Recent Advances in 3D Imaging, Modeling, and Reconstruction, edited by A. Voulodimos, A. Doulamis, IGI Global, 2020.


Modern Deep Learning techniques have proven that they have the capacity to be successful in a wide area of domains and tasks, including applications related to 3D and 2D images. However, their quality is directly dependent on the quality and quantity of the data with which models are trained. This fact becomes increasingly relevant as the capacity of deep learning models increases, and data availability becomes the most significant obstacle with regard to their application. To counter this issue various techniques are utilized, including data augmentation. Data augmentation refers to the practice of expanding the original dataset with artificially created samples, in order to train a model with data interpretations that will hopefully equip it with better generalization properties. With regard to data augmentation, one approach that has been found to show great promise, are Generative Adversarial Networks (GANs). Unlike other methods that apply domain-agnostic transformations on the original data to produce new samples (e.g. noise, rotation, flip etc.), the GAN's objective is specifically to produce diverse samples that belong to a given data distribution. Taking advantage of this property, a multitude of GAN architectures has been leveraged for data augmentation applications. The subject of this paper is to review and organize implementations of this approach on 3D and 2D imagery, examine the methods that were used, and survey the areas in which they were applied.