OmniHands
Robust 4D hand mesh recovery for interactive hands
Project Page

OmniHands: Towards Robust 4D Hand Mesh Recovery via A Versatile Transformer

OmniHands robustly recovers interactive hand meshes and their relative motion from monocular inputs, while generalizing to complex interactions and challenging multi-view scenarios.

Dixuan Lin1,2, Yuxiang Zhang2, Mengcheng Li2, Qi Yan3, Qianying Wang3, Yebin Liu2, Wei Jing3, Hongwen Zhang1*
1Beijing Normal University    2Tsinghua University    3Lenovo
Code Paper BibTeX
Results

Single Image Input

Robust reconstruction results on challenging monocular inputs, including occlusion, unusual pose, and difficult interaction configurations.

Single image result 1
Single image result 2
Single image result 3

OmniHands demonstrates robust performance in complex single-image cases.

Results

Complex Interactions

Reconstruction results on dense hand-hand interactions with severe ambiguity, contact, and self-occlusion.

Complex interaction result 1
Complex interaction result 2
Complex interaction result 3

OmniHands remains stable and accurate in difficult interaction-heavy scenes.

Results

Multi-Cameras

Applications to multi-view setups show strong robustness under occlusion and large viewpoint changes.

Multi-camera result 1
Multi-camera result 2
Multi-camera result 3

OmniHands can handle highly occluded cases when applied to multi-view tasks.

Citation

Paper and BibTeX

If you find OmniHands useful in your research, please consider citing our paper.

https://dl.acm.org/doi/abs/10.1145/3807943 
@article{lin2026omnihands,
  title={OmniHands: Robust Motion Capture of Interactive Hands via A Versatile Transformer},
  author={Lin, Dixuan and Zhang, Yuxiang and Li, Mengcheng and Jing, Wei and Yan, Qi and Wang, Qianying and Liu, Yebin and Zhang, Hongwen},
  journal={ACM Transactions on Graphics},
  year={2026},
  publisher={ACM New York, NY}
}