DeformSplat

This is the official code for Rigidity-Aware 3D Gaussian Deformation from a Single Image represented on SIGGRAPH ASIA 25.

1. Enviroment setup

conda create -y --name deformsplat python=3.10
conda activate deformsplat

pip install torch==2.4.1+cu118 torchvision --index-url https://download.pytorch.org/whl/cu118
pip install ninja numpy jaxtyping rich
pip install gsplat --index-url https://docs.gsplat.studio/whl/pt24cu118
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl
pip install --no-build-isolation git+https://github.com/rahul-goel/fused-ssim@328dc9836f513d00c4b5bc38fe30478b4435cbb5
pip install -r requirements.txt

(Optional) We recommand to login with wandb to visualize result.

wandb login

2. Run with Diva360 data

2.1 Data Setup

Please download preprocessed diva360 data:

mkdir data
cd data

# diva360 data (~200MB)
gdown 1lRq22wxogt8x8TdCdd6AMZRieElrjO1w 
unzip diva360_processed.zip

cd ..

Then your file structure will be:

data
|-- diva360_processed
|   |-- penguin_0217
|   |-- penguin_0239
|   |-- blue_car_0142
|   |-- ...

(Optional) To test arbitrary frames other than those used in our paper, you can download the original dataset (~1.8TB).

Download Diva360 (~1.8TB)

Download the data from the link below and arrange the files as shown.

• Diva360 Download Link

data
|-- diva360
|   |-- penguin
|   |   |-- frames_1
|   |   |-- segmented_gt
|   |   |-- segmented_ngp
|   |   |-- transforms_circle.json
|   |   |-- ...
|   |   
|   |-- blue_car
|   |-- ...

After Download original link, you can preprocess the data with run command (see 2.2).

2.2 Run

Please run the code below to test penguin example:

#                             GPU  object_name  frame_from  frame_to cam_idx
bash scripts/deform_diva360.sh 0   penguin      0217        0239     00     

If you want to run all Diva360 data from our paper, please run below:

bash scripts/run_all_diva360.sh

For executing other frames from our paper, please download the original data (see 2.1) and change frame_from, frame_to, and cam_idx.

3. Run with DFA data

3.1 Data Setup

Please download preprocessed DFA data:

cd data

gdown 1Qdu0s-gogtQKyAa3BZ8BrDMCk8kdzzCf
unzip DFA_processed.zip

cd ..

Then your file structure will be:

data
|-- DFA_processed
|   |-- beagle_dog(s1)
|   |-- cat(walk_final)
|   |-- duck(walk)
|   |-- lion(Walk)
|   |-- ...

(Optional) To test arbitrary frames other than those used in our paper, you can download the original dataset (~200GB).

Download original DFA (~200GB)

Download the data from the link below and arrange the files as shown.

• DFA Download Link

data
|-- dfa
|   |-- beagle_dog
|   |-- bear
|   |-- cat
|   |-- duck
|   |-- ...

After Download original link, you can preprocess the data with run command (see 3.2).

3.2 Run

Please run the code below to test wolf example:

#                          GPU       object_name     frame_from  frame_to cam_idx
bash scripts/deform_dfa.sh ${GPU}   "wolf(Howling)"  10          60       24

If you want to run all DFA data from our paper, please run below:

bash scripts/run_all_dfa.sh

For executing other frames from our paper, you can change frame_from, frame_to, and cam_idx.

🙏 Acknowledgements

This work builds upon the fantastic research and open-source contributions from the community. We extend our sincere thanks to the authors of the following projects:

• gsplat
• RoMA
• torch-splat
• Diva360
• Artemis

📜 Citation

If you find this work helpful, please consider citing our paper:

@misc{kim2025deformsplat,
    title={Rigidity-Aware 3D Gaussian Deformation from a Single Image}, 
    author={Jinhyeok Kim and Jaehun Bang and Seunghyun Seo and Kyungdon Joo},
    year={2025},
    eprint={2509.22222},
    archivePrefix={arXiv},
    primaryClass={cs.GR},
    url={https://arxiv.org/abs/2509.22222}, 
}