PhySIC: Physically Plausible 3D Human-Scene Interaction and Contact from a Single Image

Pradyumna Yalandur Muralidhar^★, Yuxuan Xue^★†, Xianghui Xie, Margaret Kostyrko, Gerard Pons-Moll

https://github.com/user-attachments/assets/86105b9b-ff49-41de-8fd8-dbf25e3e0b26

Getting Started

Installation

Please follow the instructions in INSTALL.md to set up the environment.

Downloading Data

Please run the following script to download the required data:

bash fetch_data.sh

This requires access to SMPL, SMPL-X, AGORA, and CameraHMR. Please enter the credentials when prompted.

Running the Code

To run the code, please copy images to the images/ directory. Then, execute the following command:

python run_optimizer.py

This will save the results in the outputs/ directory.

Roadmap

• Demo code release
• Evaluation code

Acknowledgements

This code is built on top of many great open-source projects. We would like to thank the authors of the following repositories:

• OmniEraser for the image inpainting code.
• MoGe for affine-invariant depth estimation.
• CameraHMR for the initial human mesh estimation.
• WiLoR/WiLoR-mini for initial hand pose estimation.
• DECO and PROX for contact estimation and static contacts.
• DepthPro for metric depth estimation.
• HSfM and ViTPose for the ViTPose inference code and models.
• LangSAM, GroundingDINO, and Segment-Anything for the segmentation code and models.
• SMPLFitter/NLF for the SMPL-to-SMPL-X converter.

and many others.

Citation

If you find our work useful, please cite:

@inproceedings{ym2025physic,
  author    = {Yalandur Muralidhar, Pradyumna and Xue, Yuxuan and Xie, Xianghui and Kostyrko, Margaret and Pons-Moll, Gerard},
  title     = {PhySIC: Physically Plausible 3D Human-Scene Interaction and Contact from a Single Image},
  journal   = {SIGGRAPH Asia 2025 Conference Papers},
  year      = {2025},
}