GolfMate (AI-powered Golf Swing Analyzer) — Catapult Hackathon (Best AI-Driven Solution)
Built a custom dataset pipeline because no labeled golf swing dataset was available online, using raw golf swing videos as the starting data. Applied MediaPipe to extract pose landmarks from swing footage and used a synthetic shuffling technique to create mixed-quality swing sequences for training data augmentation. Prepared structured temporal landmark sequences for model development and evaluation without relying on an existing labeled dataset. • Created/curated a dataset from raw swing video sources • Extracted pose landmark features using MediaPipe • Augmented dataset diversity with synthetic shuffling • Supported GRU model training on temporal landmark data