---

TL;DR

Given a monocular video as input, our UniSH is capable of jointly reconstructing scene and human in a single forward pass, enabling effective estimation of scene geometry, camera parameters and SMPL parameters.

🛠️ Installation

We provide a sudo-free installation method that works on most Linux servers (including headless ones).

Step 1: Clone Repository

git clone https://github.com/murphylmf/UniSH.git
cd UniSH

Step 2: Create Conda Environment

This installs Python, system compilers, and OpenGL drivers.

conda env create -f environment.yml
conda activate unish

Step 3: Compile Dependencies

This script compiles PyTorch3D, MMCV, and SAM2 from source using the compilers installed in Step 2.

The environment has been tested on CUDA 12.1 and CUDA 11.8. You can specify the CUDA version by passing it as an argument to the installation script.

# Default (Auto-detect or 12.1)
bash install.sh
For CUDA 11.8
bash install.sh 11.8
For CUDA 12.1
bash install.sh 12.1

Step 4: Download SMPL Models

Please download the SMPL models and place them in the body_models folder. The directory structure should be organized as follows:

UniSH/
├── body_models/
│   └── smpl/
│       └── smpl/
│           ├── SMPL_FEMALE.pkl
│           ├── SMPL_MALE.pkl
│           └── SMPL_NEUTRAL.pkl

🚀 Quick Start (Inference)

Run Inference

Run the following command to reconstruct the scene and human from the video:

python inference.py --output_dir inference_results/example --video_path examples/example_video.mp4 

Please refer to inference.py for more information about additional parameters.

📝 Citation

If you find this code useful for your research, please consider citing our paper:

@misc{li2026unishunifyingscenehuman,
      title={UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass}, 
      author={Mengfei Li and Peng Li and Zheng Zhang and Jiahao Lu and Chengfeng Zhao and Wei Xue and Qifeng Liu and Sida Peng and Wenxiao Zhang and Wenhan Luo and Yuan Liu and Yike Guo},
      year={2026},
      eprint={2601.01222},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2601.01222}, 
}

🙏 Acknowledgements

We acknowledge the excellent contributions from the following projects:

• GVHMR
• BEDLAM
• SMPL
• VGGT
• Pi3
• MoGe2

📄 License

This project is licensed under the Apache 2.0 License. See LICENSE for details.

--- Tranlated By Open Ai Tx | Last indexed: 2026-03-24 ---