SPRig: Self-Supervised Pose-Invariant Rigging from Mesh Sequences

Ruipeng Wang^1*   Langkun Zhong^2*   Miaowei Wang³

¹University of Pennsylvania   ²The University of Hong Kong   ³The University of Edinburgh
(* Equal contribution)

---

TL;DR: Since an animated sequence represents the same underlying object, we enforce a consistency prior to fine-tune existing rigging models, enabling them to learn robust, pose-invariant rigs from abundant unlabeled data.

Comparison of our method vs Puppeteer. Our method (top, blue) yields a complete, temporally consistent skeleton with smooth, coherent skinning weights, whereas Puppeteer (bottom, red) produces an incomplete skeleton with missing hand rigging and unstable, blocky skinning.

---

🚧 Code Release

This repository is currently under construction. We are organizing the clean version of the training and inference code into the src/ directory.

Currently, the repository structure is:

• doc/: Contains the source code for the Project Page.
• src/: (Coming Soon) Will contain the official implementation.

🧪 Experimental Code (Colab & Drive)

While we clean up the code, you can access our experimental notebooks and checkpoints via the links below. These notebooks were used to run the experiments in the paper on NVIDIA A100 GPUs.

Resource	Description	Link
Google Colab	Training & Inference Notebooks
Google Drive	All Checkpoints & Sample Data	📂 Open Drive Folder

Note: The code in the Colab notebooks is raw and experimental. We are working on merging it into this repository.

---

📝 Citation

If you find our work useful for your research, please consider citing:

@misc{wang2026sprig,
      title={SPRig: Self-Supervised Pose-Invariant Rigging from Mesh Sequences}, 
      author={Ruipeng Wang and Langkun Zhong and Miaowei Wang},
      year={2026},
      eprint={2602.12740},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={[https://arxiv.org/abs/2602.12740](https://arxiv.org/abs/2602.12740)}, 
}