Spaces:
Running on Zero
Running on Zero
| title: GEM-X Motion Capture | |
| emoji: 🕺 | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: gradio | |
| app_file: app.py | |
| pinned: false | |
| license: apache-2.0 | |
| python_version: 3.12 | |
| short_description: Video to SOMA motion and Blender BVH with NVIDIA GEM-X | |
| # GEM-X Motion Capture | |
| Upload a short, single-person video and recover NVIDIA SOMA whole-body motion. | |
| Outputs: | |
| - a 77-keypoint tracking preview; | |
| - a Blender-friendly 77-joint SOMA BVH; | |
| - a 78-node BVH retaining GEM-X's virtual root; | |
| - the original GEM-X PyTorch result and a ZIP archive. | |
| The source pipeline is NVIDIA GEM-X. Model weights are downloaded from | |
| `nvidia/GEM-X` and are governed by the NVIDIA Open Model License Agreement. | |
| The application code and GEM-X source are Apache-2.0. | |
| For best results, use a 1–2 second clip with one fully visible person. The | |
| first run is slower because model assets are downloaded and cached. | |