Pose Estimation System

Build a human pose estimation system.

Project details:
- Pose type: [2D/3D]
- Single/Multi person: [SINGLE/MULTI]
- Body parts: [FULL BODY/UPPER BODY/HANDS/FACE]
- Real-time: [YES/NO]

Pose estimation requirements:
1. Model selection:
   - OpenPose
   - MediaPipe
   - HRNet
   - ViTPose
2. Detection pipeline:
   - Person detection (if multi-person)
   - Keypoint detection
   - Tracking across frames
3. Post-processing:
   - Keypoint refinement
   - Temporal smoothing
   - Occlusion handling
4. 3D lifting (if needed):
   - 2D to 3D lifting
   - Multi-view triangulation
   - Depth estimation
5. Output formats:
   - COCO keypoints
   - Custom skeleton
   - Animation-ready
6. Evaluation:
   - PCK, OKS metrics
   - Speed benchmarking
   - Failure case analysis
7. Applications:
   - Action recognition
   - Gesture control
   - Motion capture

Optimize for target frame rate.