What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs

3D-aware Generative Adversarial Networks (GANs) have shown remarkable progress in learning to generate multi-view-consistent images and 3D geometries of scenes from collections of 2D images via neural volume rendering.