|
Download README.md from narusuzuki/AffHandGen: direct link, hf CLI and curl.
- Browser
- Download file 1.3 kB
-
https://huggingface.co/narusuzuki/AffHandGen/resolve/main/README.md
- Command line
-
hf download hf://narusuzuki/AffHandGen/README.md
-
curl -L -o README.md https://huggingface.co/narusuzuki/AffHandGen/resolve/main/README.md
1.3 kB
| license: mit | |
| tags: | |
| - 3d-hand-reconstruction | |
| - diffusion | |
| - hand-object-interaction | |
| # AffHandGen — Affordance-Guided Diffusion Priors | |
| Diffusion prior checkpoints for **"Affordance-Guided Diffusion Prior for 3D Hand | |
| Reconstruction"** (arXiv:2510.00506). | |
| The priors generate MANO hand poses conditioned on affordance text descriptions | |
| (DistilBERT) and image features (HaMeR ViT), and are used for test-time | |
| refinement of 3D hand reconstructions under occlusion. | |
| | File | Trained on | Used for | | |
| |---|---|---| | |
| | `hog/ckpt_best.pth` | HOGraspNet (S1 train) | Table 1 HOGraspNet rows | | |
| | `ho3d/ckpt_best.pth` | HO3D | Table 1 HO3D rows | | |
| Each directory also contains the exact `train.yml` used for training. | |
| ## Usage | |
| See the [code repository](https://github.com/narudesuyo/AffHandGen) — | |
| `docs/REPRODUCING.md` describes the full reproduction procedure. Place the | |
| checkpoints at: | |
| ``` | |
| output/exp/hog/uniaff_cond_20260228_v1/ckpt_best.pth | |
| output/exp/ho3d/20260227_v1/ckpt_best.pth | |
| ``` | |
| ## Citation | |
| ```bibtex | |
| @article{suzuki2025affordance, | |
| title={Affordance-Guided Diffusion Prior for 3D Hand Reconstruction}, | |
| author={Suzuki, Naru and Ohkawa, Takehiko and Banno, Tatsuro and Lee, Jihyun and Furuta, Ryosuke and Sato, Yoichi}, | |
| journal={arXiv preprint arXiv:2510.00506}, | |
| year={2025} | |
| } | |
| ``` | |