@mastersthesis {xCasals19, title = {Synthesis of acne images for data augmentation with generative adversarial networks}, year = {2019}, abstract = {

Generative Adversarial Networks (GANs) are deep learning architectures known for their usefulness on synthesizing new images. Conditioned image generation or the synthesis of super-resolution images are some of their main uses, but they are also helpful when tackling particular image classification and segmentation problems. The latter application is the motivation for the work presented in this document.

This work studies the synthesis of acne images for data augmentation, a procedure validated using said synthetic images to tackle an image classification problem.

The main challenge will be to work around the instability in the training of GANs. Therefore, different known solutions will be implemented in order to overcome this problem.

}, author = {Roger Casals}, editor = {Ver{\'o}nica Vilaplana} }