A Geometric and Photometric Exploration of GAN and Diffusion Synthesized Faces

Classic computer-generated imagery is produced by modeling 3D scene geometry, the surrounding illumination, and a virtual camera. As a result, rendered images accurately capture the geometry and physics of natural scenes. In contrast, AI-generated imagery is produced by learning the statistical distribution of natural scenes from a large set of real images. Without an explicit 3D model of the world, we wondered how accurately synthesized content captures the 3D geometric and photometric properties of natural scenes. From a diverse set of real, GAN- and diffusion-synthesized faces, we estimate a 3D geometric model of the face, from which we estimate the surrounding 3D photometric environment. We also analyze 2D facial features - eyes and mouth - that have been traditionally difficult to accurately render. Using these models, we provide a quantitative analysis of the 3D and 2D realism of synthesized faces.