GaussianArt: Unified Modeling of Geometry and Motion for Articulated Objects (3DV 2026)

Licheng Shen✶^1,2, Saining Zhang ✶^1,2,3, Honghan Li ✶^1,2,3, Peilin Yang^1,4, Zihao Huang^1,5, Zongzheng Zhang^1,2, Hao Zhao†^2,1

✶ indicates equal contribution † corresponding author

¹Beijing Academy of Artificial Intelligence ²Institute for AI Industry Research(AIR), Tsinghua University ³Nanyang Technological University ⁴Beijing Institute of Technology ⁵Huazhong University of Science and Technology

Website | Arxiv | Data

Environment Setup

Please follow these steps to setup the environment:

git clone https://github.com/shenlc19/GaussianArt --recursive
cd GaussianArt

# create and initialize conda environment

conda create -n gaussianart python=3.10
conda activate gaussianart

# install pytorch
pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116

# install dependencies: submodules
pip install -r requirements.txt
pip install submodules/simple-knn 
pip install submodules/art-diff-gaussian-rasterization

Pytorch3d(v0.7.5) is also required. Please clone the repository, checkout v0.7.5 and follow the instructions to install.

Dataset Preparation

Download the dataset from link, save the compressed files to ./data and decompress them. Each instance follows the structure below:

 ./data
    ├── model_id
    │   ├── start
    │   ├── end
    │   ├── gt
    │   ├── transforms_test_end.json
    │   ├── transforms_test_start.json
    │   ├── transforms_test.json
    │   ├── transforms_train_end.json
    │   ├── transforms_train_start.json
    │   ├── transforms_train.json
    └── ...

Training

Use model_id as a parameter to run the following command:

python run.py --model_id {model_id}

The training process includes depth-semantic initialization and motion-appearance joint optimization. The results will be saved to output/MPArt-90/{model_id}.

Evaluation

Evaluate the motion axis prediction by running the following command:

python eval_axis.py -m output/MPArt-90/{model_id}

Render

Run the following command to render the reconstructed articulated object and visualize the part-level motion:

python render_video.py -m output/MPArt-90/{model_id}

Citation

If you find our paper and/or code helpful, please consider citing:

@misc{shen2025gaussianartunifiedmodelinggeometry,
      title={GaussianArt: Unified Modeling of Geometry and Motion for Articulated Objects}, 
      author={Licheng Shen and Saining Zhang and Honghan Li and Peilin Yang and Zihao Huang and Zongzheng Zhang and Hao Zhao},
      year={2025},
      eprint={2508.14891},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2508.14891}, 
}

Acknowledgement

We acknowledge the authors of 3DGS, ArtGS, DigitalTwinArt for making their outstanding projects publicly available.