[CVPR 2024 Highlight (11.9%)] LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching

Yixun Liang$^{\color{red}{*}}$ Xin Yang$^{\color{red}{*}}$, Jiantao Lin, Haodong Li, Xiaogang Xu, Yingcong Chen$^{**}$

$\color{red}{*}$: Equal contribution. **: Corresponding author.

Paper PDF (Arxiv) | Project Page (Coming Soon) | Gradio Demo

The paper is now accepted to CVPR 2024 as poster (highlight, 11.9%).

---

:tv: Video

🎏 Abstract

We present a text-to-3D generation framework, named the LucidDreamer, to distill high-fidelity textures and shapes from pretrained 2D diffusion models.

CLICK for the full abstract

The recent advancements in text-to-3D generation mark a significant milestone in generative models, unlocking new possibilities for creating imaginative 3D assets across various real-world scenarios. While recent advancements in text-to-3D generation have shown promise, they often fall short in rendering detailed and high-quality 3D models. This problem is especially prevalent as many methods base themselves on Score Distillation Sampling (SDS). This paper identifies a notable deficiency in SDS, that it brings inconsistent and low-quality updating direction for the 3D model, causing the over-smoothing effect. To address this, we propose a novel approach called Interval Score Matching (ISM). ISM employs deterministic diffusing trajectories and utilizes interval-based score matching to counteract over-smoothing. Furthermore, we incorporate 3D Gaussian Splatting into our text-to-3D generation pipeline. Extensive experiments show that our model largely outperforms the state-of-the-art in quality and training efficiency.

🔧 Training Instructions

Our code is now released! Please refer to this link for detailed training instructions.

🤗 Gradio Demo

We are currently building an online demo of LucidDreamer with Gradio, you can check it out by clicking this link. It is still under development, and the service might not be available from time to time.

🚧 Todo

• Release the basic training codes
• Release the guidance documents
• Release the training codes for more applications

📍 Citation

@article{EnVision2023luciddreamer,
  title={Luciddreamer: Towards high-fidelity text-to-3d generation via interval score matching},
  author={Liang, Yixun and Yang, Xin and Lin, Jiantao and Li, Haodong and Xu, Xiaogang and Chen, Yingcong},
  journal={arXiv preprint arXiv:2311.11284},
  year={2023}
}

Acknowledgement

This work is built on many amazing research works and open-source projects:

• gaussian-splatting and diff-gaussian-rasterization
• Stable-Dreamfusion
• Point-E

Thanks for their excellent work and great contribution to 3D generation area.