Motius-GVHMR / README.md
ZeyuLing's picture
Normalize Motius Pipeline.from_pretrained metadata
523cb36 verified
|
Raw
History Blame Contribute Delete
1.72 kB
---
library_name: motius
pipeline_tag: other
license: other
tags:
- monocular-motion-capture
- gvhmr
---
# Motius GVHMR
Repository-native, source-pinned [GVHMR](https://github.com/zju3dv/GVHMR) for monocular
camera-relative and world-grounded human motion recovery.
- Motius model card:
<https://github.com/ZeyuLing/Motius/blob/main/docs/model_zoo/gvhmr.md>
- Pinned source revision: `6ec3ca39336c50492c0fae65fba2fb831fc7d866`
- GVHMR checkpoint SHA-256:
`4fae7da2de388d5da3514cb27a2d003f364dacb280e9cf88972b710e589c6b91`
```python
from motius.pipelines.gvhmr import GVHMRPipeline
pipeline = GVHMRPipeline.from_pretrained(
"ZeyuLing/Motius-GVHMR",
bundle_kwargs={
"python_executable": "outputs/envs/gvhmr/bin/python",
"body_models_root": "checkpoints/body_models",
},
)
result = pipeline.infer_monocular_motion_capture(
"input.mp4",
output_root="outputs/gvhmr/run_001",
)
```
## 3DPW Test
| Coverage | MPJPE | PA-MPJPE | Acceleration |
| ---: | ---: | ---: | ---: |
| 100.00% | 62.65 mm | 47.66 mm | 5.47 m/s2 |
The source migration passes exact parity over 39 fields across tracking,
camera construction, visual features, model input, and decoded model output.
## Preview
| Clip 01 | Clip 02 | Clip 03 |
| --- | --- | --- |
| [![](assets/case_01_global.webp)](assets/case_01_global.mp4) | [![](assets/case_02_global.webp)](assets/case_02_global.mp4) | [![](assets/case_03_global.webp)](assets/case_03_global.mp4) |
This artifact is restricted to educational, research, and non-profit use.
Licensed SMPL and SMPL-X body models are not included.
## Direct Loading
```python
from motius import Pipeline
pipeline = Pipeline.from_pretrained("ZeyuLing/Motius-GVHMR")
```