DiffusionGAN3D: Boosting Text-guided 3D Generation and Domain Adaptation by Combining 3D GANs and Diffusion Priors

TL;DR

DiffusionGAN3D tackles text-guided 3D domain adaptation and text-to-avatar generation by uniting pre-trained 3D GANs with diffusion priors. It finetunes an EG3D-based generator using Score Distillation Sampling while introducing a relative distance loss $L_{dis}$ to preserve diversity and a diffusion-guided reconstruction loss $L_{diff}$ for local editing; for avatar tasks, it adds a case-specific learnable triplane $T_l$ and a multi-scale total variation loss $L_{mstv}$, capped with a progressive texture refinement stage that uses 2D diffusion for texture enhancement. The approach leverages latent search via CLIP to align prompts with a strong 3D prior and uses an adaptive texture pipeline to ensure multi-view consistency. Experiments show improved fidelity, diversity, and efficiency over prior methods in both domain adaptation and text-to-avatar tasks, with ablations confirming the contribution of each component.

Abstract

Text-guided domain adaptation and generation of 3D-aware portraits find many applications in various fields. However, due to the lack of training data and the challenges in handling the high variety of geometry and appearance, the existing methods for these tasks suffer from issues like inflexibility, instability, and low fidelity. In this paper, we propose a novel framework DiffusionGAN3D, which boosts text-guided 3D domain adaptation and generation by combining 3D GANs and diffusion priors. Specifically, we integrate the pre-trained 3D generative models (e.g., EG3D) and text-to-image diffusion models. The former provides a strong foundation for stable and high-quality avatar generation from text. And the diffusion models in turn offer powerful priors and guide the 3D generator finetuning with informative direction to achieve flexible and efficient text-guided domain adaptation. To enhance the diversity in domain adaptation and the generation capability in text-to-avatar, we introduce the relative distance loss and case-specific learnable triplane respectively. Besides, we design a progressive texture refinement module to improve the texture quality for both tasks above. Extensive experiments demonstrate that the proposed framework achieves excellent results in both domain adaptation and text-to-avatar tasks, outperforming existing methods in terms of generation quality and efficiency. The project homepage is at https://younglbw.github.io/DiffusionGAN3D-homepage/.

Figures (11)

• Figure 1: Some results of the proposed DiffusionGAN3D on different tasks.
• Figure 2: Overview of the proposed two-stage framework DiffusionGAN3D.
• Figure 3: An illustration of the relative distance loss.
• Figure 4: Visualizations of the gradient response of SDS loss at different noise levels, given the text "a man with green hair".
• Figure 5: The details of the proposed adaptive blend module.