EXAMINING PERFORMANCE OF IDINVERT GAN MODEL ON IMITATING REAL HUMAN FACES

Abstract

This study evaluates the performance of IDInvert, a variant of the generative adversarial network (GAN) models, in terms of ability to generate synthetic face images that resemble real ones, while also preserving personal identity. The main focus of the study is to investigate whether current techniques can detect the subtle differences between real and synthetic face images generated by the IDInvert model. The findings reveal that although the IDInvert model produces highly realistic facial images, they do not preserve personal identity, and they can be identified using feature extraction techniques and standard classification models. Overall, the study highlighted the potential risks of using GAN inversion models and emphasized the importance of developing more robust and secure algorithms to prevent the misuse of such technology.

-