Learning Effective NeRFs and SDFs Representations with 3D Generative Adversarial Networks for 3D Object Generation

TL;DR

The paper tackles 3D object generation by training a 3D GAN-based pipeline that learns NeRFs for high-fidelity rendering and SDFs for mesh representation. Leveraging the EG3D framework with label embeddings and color mapping, the method enables cross-taxonomy training across 216 OmniObject3D classes, using only a few images per object. A latent-to-geometry-and-texture pipeline produces tri-plane features and textures, with an SDF decoder yielding meshes and an attention-based renderer for 2D views, all trained with a composite loss and ADA. Results show competitive performance and top-3 ranking in the OmniObject3D Challenge, with ablations confirming the benefit of label embeddings and a GAN-based approach in a multi-class setting. The work demonstrates that 3D GANs remain viable and effective for high-quality 3D generation, even as diffusion models dominate many 3D tasks.

Abstract

We present a solution for 3D object generation of ICCV 2023 OmniObject3D Challenge. In recent years, 3D object generation has made great process and achieved promising results, but it remains a challenging task due to the difficulty of generating complex, textured, and high-fidelity results. To resolve this problem, we study learning effective NeRFs and SDFs representations with 3D Generative Adversarial Networks (GANs) for 3D object generation. Specifically, inspired by recent works, we use the efficient geometry-aware 3D GANs as the backbone incorporating with label embedding and color mapping, which enables to train the model on different taxonomies simultaneously. Then, through a decoder, we aggregate the resulting features to generate Neural Radiance Fields (NeRFs) based representations for rendering high-fidelity synthetic images. Meanwhile, we optimize Signed Distance Functions (SDFs) to effectively represent objects with 3D meshes. Besides, we observe that this model can be effectively trained with only a few images of each object from a variety of classes, instead of using a great number of images per object or training one model per class. With this pipeline, we can optimize an effective model for 3D object generation. This solution is among the top 3 in the ICCV 2023 OmniObject3D Challenge.

Figures (3)

• Figure 1: An overview of the framework of our solution for 3D object generation.
• Figure 2: Images rendered by our solution with label embedding.
• Figure 3: Images rendered by our solution without label embedding.