SDF-based 3D Generative Adversarial Networks for Image and Geometry Generation

This paper presents a novel approach to generating high-quality 3D models using a 3D generative adversarial network (GAN) architecture with differentiable SDF-based rendering techniques. The system utilizes a hybrid representation model comprising tri-planes and a tiny MLP, which allows for the generation of detailed features and efficient training under the supervision of low-resolution 2D images. Additionally, depth map supervision is introduced during the training phase to enhance convergence and improve shape generation and image synthesis. Experimental results demonstrate that the proposed approach outperforms existing state-of-the-art 3D generative models in terms of geometry quality and texture performance.