Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[CVPR 2024 Highlight] Code for "HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting"
| Date | Stars |
|---|---|
| 2026-07-24 | 492 |
| 2026-07-25 | 492 |
| 2026-07-28 | 492 |
| 2026-07-30 | 492 |
| 2026-08-06 | 492 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting [Xian Liu](https://alvinliu0.github.io/)<sup>1</sup>, [Xiaohang Zhan](https://xiaohangzhan.github.io/)<sup>2</sup>, [Jiaxiang Tang](https://me.kiui.moe/)<sup>3</sup>, [Ying Shan](https://scholar.google.com/citations?user=4oXBp9UAAAAJ)<sup>2</sup>, [Gang Zeng](https://scholar.google.com/citations?user=RuHyY6gAAAAJ)<sup>3</sup>, [Dahua Lin](http://dahua.site/)<sup>1</sup>, [Xihui Liu](https://xh-liu.github.io/)<sup>4</sup>, [Ziwei Liu](https://liuziwei7.github.io/)<sup>5</sup>. <sup>1</sup>CUHK <sup>2</sup>Tencent AI Lab <sup>3</sup>PKU <sup>4</sup>HKU <sup>5</sup>NTU ### [Project](https://alvinliu0.github.io/projects/HumanGaussian) | [Paper](https://alvinliu0.github.io/projects/HumanGaussian/humangaussian.pdf) | [arXiv](https://arxiv.org/abs/2311.17061) | [Demo](https://www.youtube.com/watch?v=S3djzHoqPKY) Realistic 3D human generation from text prompts is a desirable yet challenging task. Existing methods optimize 3D representations like mesh or neural fields via score distillation sampling (SDS), which suffers from inadequate fine details or excessive training time. In this paper, we propose an efficient yet effective framework, **HumanGaussian**, that generates high-quality 3D humans with fine-grained geometry and realistic appearance. Our key insight is that 3D Gaussian Splatting is an efficient renderer with periodic Gaussian shrinkage or growing, where such adaptive density control can be naturally guided by intrinsic human structures. Specifically, **1)** we first propose a *Structure-Aware SDS* that simultaneously optimizes human appearance and geometry. The multi-modal score function from both RGB and depth space is leveraged to distill the Gaussian densification and pruning process. **2)** Moreover, we devise an *Annealed Negative Prompt Guidance* by decomposing SDS into a noisier generative score and a cleaner classifier score, which well addresses the over-saturation issue. The floating artifacts are further eliminated based on Gaussian size in a prune-only phase to enhance generation smoothness. Extensive experiments demonstrate the superior efficiency and competitive quality of our framework, rendering vivid 3D humans under diverse scenarios. <img src='./content/humangaussian-teaser.gif' width=800> <img src='./content/teaser-1.png' width=800> <img src='./content/teaser-2.png' width=800> ## News * [2023-12-05] Update the real-time animation demo and codes! * [2023-11-28] Upload the paper. Release all the training codes and pretrained models! ## Installation ``` # clone the github repo git clone https://github.com/alvinliu0/HumanGaussian.git cd HumanGaussian pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118 pip install -r requirements.txt # a modified gaussian splatting (+ depth, alpha rendering) git clone --recursive https://github.com/ashawkey/diff-gaussian-rasterization pip install ./diff-gaussian-rasterization ``` ## Text Prompts Gallery The text prompts that are used for qualitative/ablation visual results demonstration are included in `prompts_gallery.txt`. ## Pretrained Models We recommend you to prepare below pre-trained models before the training process: * [SMPL-X](https://smpl-x.is.tue.mpg.de/). Please kindly refer to the SMPL-X project page for the download instructions; * [Texture-Structure Joint Model](https://mycuhk-my.sharepoint.com/:u:/g/personal/1155165198_link_cuhk_edu_hk/EYFLeeQznhRMk2OSNIt5a4EB27Vrx36Y7Nl4RbSbVGFSHQ?e=EkBNhW), which is trained on the text-image-depth pairs annotated by [MiDaS](https://github.com/isl-org/MiDaS) on the LAION dataset. After downloading the above models, you could specify the path to SMPL-X model in `system.smplx_path`, and the path to Texture-Structure Joint Model in `system.guidance
Excerpt of 9,604 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5eb95cc186046dea, topic:text-to-3d