[CVPR 2026 Highlight] PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning

PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning, Yuanhang Lei, Tao Cheng, Xingxuan Li, Boming Zhao, Siyuan Huang, Ruizhen Hu, Peter Yichen Chen, Hujun Bao, Zhaopeng Cui†

PhysSkin learns continuous skinning fields for real-time, physics-based animation across different shapes and discretizations. It combines a transformer-based shape encoder, a cross-attention field decoder, and physics-informed self-supervised objectives.

The code was tested with Python 3.11, PyTorch 2.7.1, NumPy 1.26.4, PyTorch Lightning 1.6.4, and conflictfree 0.1.8.

python -m pip install "pip<24.1"
pip install torch==2.7.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

<dataset_root>/
└── 00000001/
    └── <model_id>/
        └── models/
            └── samples/
                ├── latents.npz
                └── internal_filled.npz

• latents.npz contains latents, a float32 array of shape (1, 256, 768).
• internal_filled.npz contains points, a float32 array of shape (N, 3) in approximately [-0.5, 0.5].

The loader rescales the points to approximately [-1, 1]. Data splits are provided in data/splits, and processed examples for visualization are included in data/examples.

physskin_vis.ipynb

The notebook uses data/examples by default. A different sample or dataset can be selected with MODEL_ID and DATASET_ROOT.

DATASET_ROOT=/path/to/dataset \
GPU_IDS="0 1" \
bash scripts/train_physskin.sh

pip install -r data_process/requirements.txt

DATASET_ROOT=/path/to/dataset \
SAMPLING_PYTHON=/path/to/python \
RAYTRACING_PYTHON=/path/to/python \
MICHELANGELO_ROOT=/path/containing/Michelangelo \
ENCODER_CHECKPOINT=/path/to/shape_encoder.pth \
bash data_process/process_data.sh 00000001 0

The pretrained checkpoint is stored at checkpoints/physskin_rignet_epoch150.pt.

This project builds upon Simplicits, Michelangelo, and ConFIG. See THIRD_PARTY_NOTICES.md for details.

@InProceedings{Lei_2026_CVPR,
    author    = {Lei, Yuanhang and Cheng, Tao and Li, Xingxuan and Zhao, Boming and Huang, Siyuan and Hu, Ruizhen and Chen, Peter Yichen and Bao, Hujun and Cui, Zhaopeng},
    title     = {PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2026},
    pages     = {32357-32366}
}