FAB / README.md
noahcao's picture
Add arXiv link and citation
f2e204f verified
|
Raw History Blame Contribute Delete
2.18 kB
---
license: other
license_name: fab-weights-license
license_link: LICENSE
tags:
- human-pose-estimation
- 3d-human-reconstruction
- smpl
- human-contact
- humanoid-retargeting
---
# FAB model weights
Weights for **FAB: Fast and Accurate Body Estimation in Pixels, World Space and Scenes**.
Code, installation and usage: [github.com/noahcao/FAB](https://github.com/noahcao/FAB). [Project page](https://www.jinkuncao.com/FAB/) · [Paper (arXiv)](https://arxiv.org/abs/2611.25801)
## Download
```bash
hf download noahcao/FAB --local-dir ~/.cache/fab/weights
```
FAB reads the weights from `~/.cache/fab/weights`, `$FAB_WEIGHTS` or `--weights DIR`. SMPL, the SAM 3D Body config and
the MHR model are not included; the repository README explains how to get them.
## Contents
| Path | Contents |
|---|---|
| `fab-l/`, `fab-s/` | `model.pt` (FAB-L, the default, and FAB-S: keypoints, dense surface points, contacts, parts and the G1 head), `config.json` |
| `fab-h/` | `model.pt` (FAB-H: MHR mesh, keypoints and dense points) |
| `shared/detector.pt` | person detector |
| `shared/smpl_fitter.pt`, `shared/smpl_fitter_mesh.pt` | SMPL fitters (from dense points; from FAB-H's MHR mesh) |
| `shared/contact_video.pt` | causal video contact model (world mode) |
| `SHA256SUMS` | checksums of the model files and configs |
## Verify downloads
```bash
cd ~/.cache/fab/weights
sha256sum -c SHA256SUMS
```
On macOS, use `shasum -a 256 -c SHA256SUMS`.
## Licence
Research and non-commercial use only, under [LICENSE](LICENSE). The pose models, the SMPL fitters and the video contact
model derive from SAM 3D Body and DINOv3 and carry the SAM License and the DINOv3 License (reproduced in LICENSE).
Built with DINOv3. The detector's provenance and upstream terms are in Part 4 of LICENSE. No dataset images or the SMPL
model are included.
## Citation
```bibtex
@misc{fab2026,
title = {Fast and Accurate Body Estimation in Pixels, World Space and Scenes},
author = {Cao, Jinkun},
year = {2026},
eprint = {2611.25801},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2611.25801}
}
```