Face Components Segmentation Using Generated Facial Images

Machine learning methods require massive amounts of data in order to satisfy quality constraints, and using real data exclusively sometimes leads to limited results. This paper presents semantic face segmentation using generated face images based on a generative adversarial network. Therefore, through a convolutional neural network added as an extension of the generator, semantic segmentation labels are obtained for the generated images, with a mean F1 of 92.01% for 18 classes.