MediaPipe Edge Discriminator (Chen et al. 2025)

Automated blade edge classification pipeline using MediaPipe pose geometry and curated blade angle heuristics.

Primary tasks
Jump mechanics, Edge analysis, Element validation
Modalities
RGB video, 2D pose keypoints, Audio cues
Architecture
Rule-augmented neural classifier over MediaPipe features
Frameworks
TensorFlow
Availability
research
Maintainer
Fudan University
Released
Feb 2025

Highlights

Implementation Notes

For robustness, retrain the classifier on local rink footage to capture camera placement differences. Combine with FS-Jump3D Pose Estimator outputs when depth estimates are required for ambiguous edges. The authors share the 2025 preprint privately; reach out to the corresponding author to request access and confirm the latest heuristics.