Publication | Open Access
Efficient Geometry-aware 3D Generative Adversarial Networks
46
Citations
0
References
2021
Year
Unsupervised generation of high‑quality, multi‑view‑consistent 3D shapes from single‑view 2D photographs remains difficult because existing 3D GANs are either compute‑intensive or rely on approximations that hurt quality and consistency. This work aims to enhance both computational efficiency and image quality of 3D GANs without heavily relying on such approximations. We propose a hybrid explicit‑implicit network that decouples feature generation from neural rendering, enabling the use of efficient 2D CNN generators like StyleGAN2 to produce real‑time, high‑resolution, multi‑view‑consistent images and high‑quality 3D geometry. The method achieves state‑of‑the‑art 3D‑aware synthesis on FFHQ and AFHQ Cats, among other experiments.
Unsupervised generation of high-quality multi-view-consistent images and 3D shapes using only collections of single-view 2D photographs has been a long-standing challenge. Existing 3D GANs are either compute intensive or make approximations that are not 3D-consistent; the former limits quality and resolution of the generated images and the latter adversely affects multi-view consistency and shape quality. In this work, we improve the computational efficiency and image quality of 3D GANs without overly relying on these approximations. We introduce an expressive hybrid explicit implicit network architecture that, together with other design choices, synthesizes not only high-resolution multi-view-consistent images in real time but also produces high-quality 3D geometry. By decoupling feature generation and neural rendering, our framework is able to leverage state-of-the-art 2D CNN generators, such as StyleGAN2, and inherit their efficiency and expressiveness. We demonstrate state-of-the-art 3D-aware synthesis with FFHQ and AFHQ Cats, among other experiments.