metadata
license: other
license_name: sam-license
license_link: >-
https://huggingface.co/peabody124/Sam3dBodyEqx/blob/main/LICENSE.component.sam-3d-body
library_name: jax
pipeline_tag: image-to-3d
tags:
- human-mesh-recovery
- 3d-human-pose
- jax
- equinox
- sam-3d-body
- mhr
Sam3dBodyEqx — converted weights
Pre-converted JAX/Equinox (.eqx) weights for Sam3dBodyEqx,
a pure JAX/Equinox port of Meta's SAM 3D Body
for single-image full-body 3D human mesh recovery. No PyTorch runtime is needed for inference.
Files
| File | What it is | License |
|---|---|---|
sam3d_body.eqx |
Converted SAM 3D Body model (backbone + decoder + heads) | SAM License — see LICENSE.component.sam-3d-body |
mhr.eqx |
Converted Momentum Human Rig (MHR) forward-kinematics body model | Apache-2.0 — see LICENSE.component.mhr |
These are converted forms of Meta's released models; the upstream licenses are
preserved and redistributed here. The SAM License is maintained for
sam3d_body.eqx (no military/nuclear/espionage or ITAR use; redistributions
must include the license; patent litigation terminates the grant).
Licensing summary
- Weights (this repo): each file keeps its upstream license as in the table above.
- Inference code (the GitHub repo): original code is AGPL-3.0-or-later. The code license does not relicense these weights — they remain under SAM License / Apache-2.0 respectively.
Usage
from sam3d_body_eqx.mhr.mhr_model import MHRForwardKinematics
mhr = MHRForwardKinematics.from_pretrained(
repo_id="peabody124/Sam3dBodyEqx", filename="mhr.eqx"
)
Citations
@article{yang2025sam3dbody,
title={SAM 3D Body: Robust Full-Body Human Mesh Recovery},
author={Yang, Xitong and Kukreja, Devansh and Pinkus, Don and others},
journal={arXiv preprint}, year={2025}
}
@misc{MHR:2025,
title={MHR: Momentum Human Rig},
author={Ferguson, Aaron and Osman, Ahmed A. A. and Bescos, Berta and others},
year={2025}, eprint={2511.15586}, archivePrefix={arXiv}, primaryClass={cs.GR}
}