Instructions to use Blendcap/sam-3d-body with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- SAM 3D Body
How to use Blendcap/sam-3d-body with SAM 3D Body:
from notebook.utils import setup_sam_3d_body estimator = setup_sam_3d_body(Blendcap/sam-3d-body) outputs = estimator.process_one_image(image) rend_img = visualize_sample_together(image, outputs, estimator.faces)
- Notebooks
- Google Colab
- Kaggle
| license: other | |
| license_name: sam-license | |
| license_link: https://huggingface.co/Blendcap/sam-3d-body/blob/main/LICENSE | |
| tags: | |
| - sam-3d-body | |
| - 3d-body | |
| - performance-capture | |
| base_model: | |
| - facebook/sam-3d-body-dinov3 | |
| # SAM 3D Body (DINOv3) β BlendCap Mirror | |
| Mirror of [facebook/sam-3d-body-dinov3](https://huggingface.co/facebook/sam-3d-body-dinov3) for use by the [BlendCap](https://github.com/Arcomade/BlendCap) performance capture addon's installer. | |
| Original weights and license belong to Meta. The SAM License (see `LICENSE`) governs use of these weights. This repository is not affiliated with or endorsed by Meta. | |
| ## License β read before downloading | |
| **By downloading or using the files in this repository you accept the license terms that govern them.** Under the SAM License's own terms (Section 6), the agreement commences upon acceptance of the agreement *or access to the materials*. A copy is included in this repository as `LICENSE`; the convenience copies below keep their own licenses (`LICENSE_YOLO_AGPL.txt`, `LICENSE_MOGE_MIT.txt`). | |
| In short: commercial use is permitted royalty-free, redistribution must carry the license text along, use must comply with the license's restrictions (including its trade-control and weapons-related prohibitions), and everything is provided as-is with no warranty. | |
| ## Files | |
| ### SAM 3D Body (DINOv3) β primary | |
| | File | Size | Description | License | | |
| |---|---|---|---| | |
| | `model.ckpt` | 2.1 GB | Main SAM 3D Body checkpoint (DINOv3 backbone) | SAM License (see `LICENSE`) | | |
| | `assets/mhr_model.pt` | 696 MB | MHR (Meta Human Rig) parametric model | SAM License (see `LICENSE`) | | |
| | `model_config.yaml` | 1.5 KB | Model configuration | SAM License (see `LICENSE`) | | |
| ### Convenience copies β pinned versions of upstream models | |
| | File | Size | Source | License | | |
| |---|---|---|---| | |
| | `yolo11m-pose.pt` | 42 MB | YOLO11m-pose from Ultralytics | AGPL-3.0 (see `LICENSE_YOLO_AGPL.txt`) | | |
| | `moge-2-vits-normal.pt` | 140 MB | MoGe-2 ViT-S Normal from Microsoft | MIT (see `LICENSE_MOGE_MIT.txt`) | | |
| The SAM 3D Body files mirror the original `facebook/sam-3d-body-dinov3` repo unmodified. YOLO and MoGe copies are pinned snapshots from their respective upstream sources, included so the installer's first-run flow can pull all PyTorch weights from one place. | |
| ## Source | |
| Original repo: [facebook/sam-3d-body-dinov3](https://huggingface.co/facebook/sam-3d-body-dinov3) (gated β requires HF account + Meta license acceptance). | |
| ## Note on "Unsafe" file labels | |
| HuggingFace flags `.pt`/`.pth` files as "unsafe" by default because the pickle format CAN contain executable code. This label applies to `mhr_model.pt` and `yolo11m-pose.pt` in this repo. These files are bit-for-bit identical to their upstream sources ([Meta's SAM 3D Body](https://huggingface.co/facebook/sam-3d-body-dinov3/tree/main/assets) and [Ultralytics YOLO11](https://huggingface.co/Ultralytics/YOLO11/tree/daa93d8d54f3f3463965b19c41e62e94a3f377d1)) where the same warning appears. No code modifications. | |