2.9 KiB
Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging
V0.1, Introduced By GAP Lab from CUHKSZ and Game-AIGC Team from ByteDance
Hi3DGen target at generating high-fidelity 3D geometry from images using normal maps as an intermediate representation. The framework addresses limitations in existing methods that struggle to reproduce fine-grained geometric details from 2D inputs.
News
[April 13, 2025] 🚀 Hi3DGen v0.1 Released! 🔥🔥🔥 Hi3DGen-MV is still under training, we expect to release it by April 20th. Model Offloading for compatible 8 GB GPU memory is expected by April 17th.
Installation
Clone the repo:
git clone --recursive https://github.com/Stable-X/Hi3DGen.git
cd Hi3DGen
Create a conda environment (optional):
conda create -n stablex python=3.10
conda activate stablex
Install dependencies:
# pytorch (select correct CUDA version)
pip install torch==2.4.0 torchvision==0.19.0 --index-url https://download.pytorch.org/whl/{your-cuda-version}
pip install spconv-cu{your-cuda-version}==2.3.6 xformers==0.0.27.post2
# other dependencies
pip install -r requirements.txt
Local Demo 🤗
Run by:
python app.py
License
The model and code of Hi3DGen are adapted from Trellis, which are licensed under the MIT License. While the original Trellis is MIT licensed, we have specifically removed its dependencies on certain NVIDIA libraries (kaolin, nvdiffrast, flexicube) to ensure this adapted version can be used commercially. Hi3DGen itself is distributed under the MIT License.
Citation
If you find this work helpful, please consider citing our paper:
@article{ye2025hi3dgen,
title={Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging},
author={Ye, Chongjie and Wu, Yushuang and Lu, Ziteng and Chang, Jiahao and Guo, Xiaoyang and Zhou, Jiaqing and Zhao, Hao and Han, Xiaoguang},
journal={arXiv preprint arXiv:2503.22236},
year={2025}
}