Smart Gym - Exercise Tracker

Comprehensive computer vision pipeline for fitness analysis. Identifies people, classifies exercises, counts repetitions, and produces structured analysis. Built with MediaPipe pose estimation, InsightFace face recognition, and scikit-learn classification.

Classifies 22 exercise types with 96.1% cross-validation accuracy (96.9% holdout). Uses peak-valley detection for exercise-independent rep counting.

Key Features

• 22 exercise classes classified (96.1% CV accuracy)
• InsightFace (buffalo_l) for person identification
• MediaPipe PoseLandmarker (multi-person up to 3)
• Peak-valley rep counting algorithm
• Label stabilization (EMA + Hysteresis + Min-hold)
• Annotated output video with skeleton overlay
• Structured JSON output with rep timestamps
• Supports: bench press, squat, deadlift, pull-ups, lat pulldown, and 17 more exercises