SDFit: 3D Object Pose and Shape by Fitting a Morphable SDF to a Single Image

Dimitrije Antić¹    Georgios Paschalidis¹    Shashank Tripathi²   
Theo Gevers¹    Sai Kumar Dwivedi²    Dimitrios Tzionas^1
1 University of Amsterdam, The Netherlands     
² Max Planck Institute for Intelligent Systems, Tübingen, Germany

ICCV 2025

  

🔗 Setup • Run • Citation • Status

---

🔧 Setup

To clone the repository, run:

git clone --recurse-submodules https://github.com/anticdimi/sdfit.git

To use our method, you need to download pretrained Deep Implicit Templates checkpoints from here, and place them in the ./pretrained folder.

export CONDA_CMD=mamba # (or conda if you use conda)
$CONDA_CMD create -n sdfit_env python=3.10 -y
$CONDA_CMD activate sdfit_env

$CONDA_CMD install -c conda-forge gcc_linux-64=12.2.0 gxx_linux-64=12.2.0 ffmpeg x264 -y

pip install torch==2.3.0+cu121 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

wget https://anaconda.org/pytorch3d/pytorch3d/0.7.8/download/linux-64/pytorch3d-0.7.8-py310_cu121_pyt231.tar.bz2
$CONDA_CMD install pytorch3d-0.7.8-py310_cu121_pyt231.tar.bz2
rm -rf pytorch3d-0.7.8-py310_cu121_pyt231.tar.bz2

python assets/test_torch_p3d_install.py

pip install opencv-python tqdm pre-commit scikit-learn iopath omegaconf plyfile scikit-image tensorboardX h5py tensorboard trimesh matplotlib accelerate huggingface_hub==0.23.0 diffusers==0.21.4 xformers==0.0.26.post1 transformers==4.26.1 git+https://github.com/anticdimi/videoio.git git+https://github.com/NVlabs/nvdiffrast.git

📋 Code Release Status

We will release our code in two stages:

• Demo
• Full Framework w/ detailed instructions about steps (est. December)

🚀 Running SDFit

For running the demo, download the demo example from here, and place it under ${YOUR_DEMO_PATH}.

🧪 Demo

After setting up your environment, and downloading the demo example, run:

python -m sdfit cfg/default.yaml data.input_path=${YOUR_DEMO_PATH}

📚 Citation

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

@inproceedings{antic2025sdfit,
    title     = {{SDFit}: {3D} Object Pose and Shape by Fitting a Morphable {SDF} to a Single Image},
    author    = {Anti\'{c}, Dimitrije and Paschalidis, Georgios and Tripathi, Shashank and Gevers, Theo and Dwivedi, Sai Kumar and Tzionas, Dimitrios},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
}

🙏 Acknowledgements

Our method builds upon prior open-source efforts. We thank the authors for releasing their code and models: Deep Implicit Templates, nvdiffrast, FlexiCubes, OpenShape, and Diff3F.