Ran a small benchmark last week on SwingVision (v10.8) with an iPhone 15 at 1080p/60, baseline camera 1.25 m high, 6.5 m behind the line, comparing against hand-tagged data. Over 3 college duals (287 points), depth estimates showed about 0.42 m RMSE, serve-type recall 93% with 8% confusion between kick and slice, and rally-length was spot on; the export-to-CSV workflow was fast. Has anyone found calibration tweaks or a companion tool that improves depth and spin proxies for live decision-making, or is Playsight/Kinovea still the more reliable route courtside?
At “1080p/60”, raising baseline cam from “1.25 m” to about 1.6 m shaved about 0.1 m RMSE; still mixes kick/slice sometimes.
Lock the camera to the 1x lens and bump digital zoom to about 1.2x; that shaved about 0.1–0.15 m off depth for me by avoiding ultra‑wide distortion — small caveat: you’ll likely need to move the tripod back a bit to keep both baselines in frame, @harrisonT34.