Production sports computer vision at KhiladiPro. The stack spans annotation infrastructure, ball & shuttle tracking, smash biomechanics, audio event detection, and the benchmarking methodology that gates every model that ships. Click any card for the architecture, the measured numbers, and the honest limitations.
A model-in-the-loop annotation platform spanning six sport-specific surfaces across cricket and badminton. Not a labelling tool — a data-flywheel where every keystroke produces a prediction/correction delta that feeds the next training cycle. Measured by how much the models improve because of what it labels.
Production shuttle tracking for live badminton broadcasts. 93% accuracy on held-out clips from broadcast + KourtEye footage — lifted from 80% via flywheel-driven retraining on annotator-corrected data. Two variants: TrackNetV5 (heavy, production accuracy) and TrackNetV5Lite (INT8, on-device, 69.5 FPS).
Badminton smash biomechanics on 240 FPS mobile capture. Dual-output pipeline that classifies the smash into 6 kinematic phases at 95% and measures shuttle exit speed at 90% vs ground truth. MediaPipe Pose for the body, TrackNet for the shuttle, a Savitzky-Golay denoiser between them — coach-grade biomechanics without a mocap lab.
Bat-ball contact detection from audio, not video. The only event-detection method in the production stack that survives noisy training environments where visual methods fail — nets, occlusion, motion blur. ~85% production accuracy, fine-tune peaked at 86.36%, built in a 9-day arc with a memorable annotation-export bug in the middle.
A model isn't shipped because it works — it's shipped because we know how well it works. Cross-sport benchmarking across three production families: cricket ball tracking, shuttle tracking, badminton player classification. Held-out clips, held-out venues, composite scoring across accuracy, latency, and pixel error. The evaluation rigor is the deliverable.
Temporal clip classification for cricket and badminton — automated shot tagging that eliminates manual breakdown work and enables tactical analysis at scale. Sport-specific class hierarchies built with domain experts. Detailed writeup coming.