Advancing AI Video Analysis for Figure Skating
FSRI unites researchers, coaches, and skaters to build open, reliable tools for understanding jumps, spins, edges, and program components.
What We Focus On
We review state-of-the-art figure skating models and datasets, advocating for transparent, unbiased evaluations the community can trust.
Our small team prioritizes published work with the greatest impact on athlete safety and fair scoring. We do not advance every subfield ourselves, but we gather evidence across leading releases so advocates can see where the science stands today.
Element detection & parsing
We audit headline detectors and parsers—often powered by convolutional and transformer backbones—to understand how reliably they surface jumps, spins, and choreo sequences that officials and advocates review.
Jump mechanics & spin analysis
We compare claims about take-off edges, air time, and centering against pose-derived evidence, publishing notes that clarify where the literature agrees, disagrees, or still lacks unbiased evaluation.
Coach tools & open benchmarks
We translate lab findings into practical dashboards, briefs, and open protocols so coaches, skaters, and policymakers can pressure-test systems without needing proprietary infrastructure.
Featured Models
Production-ready backbones and research releases powering pose, detection, and scoring stacks.
MediaPipe Edge Discriminator (Chen et al. 2025)
Automated blade edge classification pipeline using MediaPipe pose geometry and curated blade angle heuristics.
Hybrid Jump Dynamics (Tanaka et al. 2023)
Sensor-fusion pipeline combining monocular broadcast video with body-mounted IMUs to recover jump rotations and airtime.
PoseFormerV2
Transformer-based 3D pose estimation model with strong temporal reasoning, ideal for jump and spin kinematics.
Datasets We Track
Working with open and request-only corpora to benchmark figure skating systems.
YourSkatingCoach
Fan-sourced practice footage with consented athlete feedback launched in 2024 for grassroots coaching analytics.
FSBench / FSAnno
Unified annotation benchmark combining historic broadcast archives with modern call sheets for reproducible evaluation.
SkatingVerse Challenge
Community benchmark released for the 2024 SkatingVerse Challenge covering detection, program breakdown, and responsible score prediction.
Get Involved
We welcome collaborators across research, coaching, and the skating community. Share datasets, validate models, or help define responsible evaluation benchmarks.
- Researchers: model design, evaluation, reproducibility.
- Coaches: domain guidance, practical validation, workflow feedback.
- Skaters & rinks: data partnerships with consent and safety commitments.
- Supporters: open science grants and sponsorships.
Email board@figureskate.ai or say hello on @FigureSkateAI.