A Lightweight, WGPU-based Real-time Visualization Framework for 3D Human Motion in Python.
PyMotionViewer fills the gap between simple 2D plots (Matplotlib) and heavy game engines (Unity/UE) for computer vision researchers. It enables high-fidelity, real-time rendering of 3D human motion (from MediaPipe, SMPL, WHAM, etc.) directly in your Python environment.
Mapping 2D/3D sparse landmarks from a single image to a full Mixamo character skeleton using our analytical retargeting algorithm.
| Input Source | MediaPipe Detection | PyMotionViewer Result |
|---|---|---|
 |
 |
 |
| Original Image | Pose Landmarks | Real-time WGPU Render |
Visualizing complex motion data (PKL files) generated by state-of-the-art algorithms like WHAM.
| Original Video | Algorithm Overlay (SMPL) | PyMotionViewer Result |
|---|---|---|
 |
 |
 |
| Input Video | WHAM Mesh Overlay | Retargeted on Mixamo Character |
- 🚀 Lightweight & Fast: Built on
pygfx(WGPU), offering next-gen graphics performance without the overhead of Unity or Blender. - 🐍 Pure Python: seamlessly integrates into your existing PyTorch/TensorFlow research workflow.
- ☠️ Universal Retargeting: Features a custom vector-algebra-based retargeting engine that maps sparse landmarks (MediaPipe) or rotation matrices (SMPL) to standard bone structures.
- 🛠 Easy to Extend: decoupled architecture allows easy integration with HMR2.0, 4D-Humans, or custom motion datasets.
git clone [https://github.com/fangbo234/PyMotionViewer.git](https://github.com/fangbo234/PyMotionViewer.git)
cd PyMotionViewer
# Install dependencies
pip install -r requirements.txt