Thermal Face Generation Using StyleGAN

This article proposes the use of generative adversarial networks (GANs) via StyleGAN2 to create high-quality synthetic thermal images and obtain training data build face recognition models using deep learning. We employed different variants StyleGAN2, incorporating new improved version StyleGAN that uses adaptive discriminator augmentation (ADA). In addition, three databases from literature were train a detector based on YOLOv3 its variants, evaluating metrics. The database was built GANSpace manipulate intermediate latent space w with characteristics, such as eyeglasses, rotation, beards, etc. carried out 6 pretrained learning for validate our database, obtaining 99.98% accuracy classifying images.