Background
CFA (Calisthenic Form Analyzer) evaluates calisthenics exercise form — planche, front lever, back lever, handstand — from a short uploaded video.
How it works
- Extracts frames at 2 FPS and runs MediaPipe pose detection over each frame
- Computes geometric metrics (joint angles, alignment, lean) against calibration models per movement
- Sends the computed metrics to an LLM (configurable: Claude, OpenAI, or Gemini) to generate personalized coaching feedback
- Visualizes results with skeleton overlays and score trend charts
Tech stack
- Backend: FastAPI, SQLAlchemy, MediaPipe, OpenCV, NumPy
- Frontend: React, TypeScript, Recharts
- Storage: SQLite
The metric configuration is designed to be extensible, so new movements can be added without touching the core analysis pipeline.