Lux3D Pi3 Canonical Weights

This repository contains the canonicalized safetensors export of the Pi3 checkpoint used by the Lux3D runtime.

Model Details

• Model family: pi3
• Canonical file: model.safetensors
• Tensor count: 1210
• Primary runtime target: Lux3D runtime
• Typical output: affine-invariant point cloud export (.ply)

Included Files

• model.safetensors
• resolved_config.json
• manifest.json
• checksums.json

Usage With Lux3D

The Lux3D CLI can validate and use the package once it is installed into your local model asset layout.

cargo run -p lux3d-cli -- inspect --repo-root <runtime-root> pi3
cargo run -p lux3d-cli -- run --repo-root <runtime-root> pi3 --source <input-sequence> --output <output-file.ply>

Provenance

• Upstream source model: yyfz233/Pi3
• Canonicalization flow: tools/python_baseline/normalize_weights.py

The exact source checksum and canonical checksum set are recorded in manifest.json and checksums.json.

Intended Use

• image or video to 3D reconstruction with Lux3D
• point cloud export workflows based on the Pi3 contract
• reproducible runtime validation

Limitations

• Verified Lux3D runtime inference currently assumes CUDA.
• This package is for the base pi3 family, not pi3x.
• The repository contains canonicalized runtime artifacts, not the full upstream training or demo stack.

License

• Upstream model page: yyfz233/Pi3
• License file in this repository: LICENSE

Review upstream terms before redistribution or commercial use.

Citation

@article{wang2025pi,
  title={$\pi^3$: Scalable Permutation-Equivariant Visual Geometry Learning},
  author={Wang, Yifan and Zhou, Jianjun and Zhu, Haoyi and Chang, Wenzheng and Zhou, Yang and Li, Zizun and Chen, Junyi and Pang, Jiangmiao and Shen, Chunhua and He, Tong},
  journal={arXiv preprint arXiv:2507.13347},
  year={2025}
}