Downloads · 30 days
0
miburi-release-checkpoints
m-hamza-mughal/miburi-release-checkpoints
miburi-release-checkpoints is a machine learning model from m-hamza-mughal. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as cc-by-nc-4.0.
Companion model weights for MIBURI: Towards Expressive Interactive Gesture Synthesis (CVPR 2026).
Access
Public
Updated Jun 4, 2026
Repo size
3.6 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors1.8 GB · 100%
At a glance
- License
- cc-by-nc-4.0
- Access
- Public
- Created
- Jun 3, 2026
- Updated
- Jun 4, 2026
- SHA
- 6541cff5
Spaces
Model card
From the Hugging Face model README
MIBURI release checkpoints
Companion model weights for MIBURI: Towards Expressive Interactive Gesture Synthesis (CVPR 2026).
- Paper: https://arxiv.org/abs/2603.03282
- Project page: https://vcai.mpi-inf.mpg.de/projects/MIBURI/
- Code: https://github.com/m-hamza-mughal/miburi
These weights are not used standalone — they are loaded by the MIBURI demo server from the code repository. The intended way to fetch them is:
git clone https://github.com/m-hamza-mughal/miburi
cd miburi
pip install .
miburi-download-assets # https://huggingface.co/m-hamza-mughal/miburi-release-assets
miburi-download-checkpoints # this repo, pulled into experiments/
miburi-demo
Contents
8 directories under experiments/, two release tracks × {3 codecs + 1
Gesture LM}:
| Dir | Role |
|---|---|
demoexp_release_uppercodec/ | Upper body + hands codec |
demoexp_release_lowercodec/ | Lower body + trans-velocity codec |
demoexp_release_facecodec/ | Face / expression codec |
demoexp_release_gtdm3_goodspk/ | GTDM3 LM, BEATX 4-speaker subset |
allspk_release_uppercodec/ | Upper codec, all-speaker |
allspk_release_lowertranscodec/ | Lower codec, all-speaker |
allspk_release_facecodec/ | Face codec, all-speaker |
allspk_release_gtdm3_exp/ | GTDM3 LM, all 23 BEATX speakers |
Each codec dir holds config.yaml + a single last_<NNN>.safetensors. The
GTDM3 dirs additionally hold speaker_id_to_index.json and
tokens_idle_slices/{idle_upper,idle_face}_tokens_slices.npz that the demo
server reads at startup.
License
These checkpoints are licensed under CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/), the same license that governs the MIBURI source code. Research and academic use is welcome; commercial use is not permitted. If you build on this work, please cite:
Mughal, M. Hamza; Dabral, Rishabh; Demberg, Vera; Theobalt, Christian. MIBURI: Towards Expressive Interactive Gesture Synthesis. CVPR 2026. https://arxiv.org/abs/2603.03282
Third-party licenses
The MIBURI demo composes these weights with third-party code + assets, each governed by their own license. By using the demo you agree to honor these as well:
- SMPL-X model (loaded by the demo for SMPL-X → vertex decoding;
shipped in the companion
miburi-release-assetsrepo) — Max Planck Society, non-commercial use only. See https://smpl-x.is.tue.mpg.de/modellicense.html. - Moshi 7B + Mimi (fetched automatically by the demo from https://huggingface.co/kyutai/moshiko-pytorch-bf16 and https://huggingface.co/kyutai/moshi-artifacts) — CC-BY 4.0 (Kyutai).
- Mixamo characters (companion
miburi-release-assetsrepo) — Adobe Mixamo terms of use: https://helpx.adobe.com/creative-cloud/faq/mixamo-faq.html.
- License
- cc-by-nc-4.0
- Created
- Jun 3, 2026
- Updated
- Jun 4, 2026