Computer Vision, Not Just for Breakfast Anymore
summary
Walks through concrete iOS code using Vision framework and Create ML to build a hands‑free pinch game: a hand pose detector that reports up to 21 points, face rectangle detection, VNTrackObjectRequest tracking to avoid double‑scoring, and featureprint comparisons. Demonstrates a pumpkin classifier exported as a 17 KB classifier, shows matching a TV frame to a movie (Spider‑Man at 1:11:06), and publishes the app code on GitHub. Works end‑to‑end in demo; expect implementation details and tradeoffs rather than high‑level theory.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:25Crushing heads game concept
- 8:07Hand pose detection
- 10:34Face detection for crushing
- 13:23Preventing duplicate crushes
- 19:27Object tracking for crushing
- 21:57Pumpkin classifier training
- 25:07Core ML classifiers
- 27:14App components summary
- 28:40Ethical data collection Q&A
- 31:08Vision model limitations Q&A
related talks
45mMobiconf 2018: Meghan Kane, Novoda
41mMaking Android Apps with Intelligence
40mNavigating the Machine Learning Landscape for Apple Devices
55mCore ML: Vision
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
42mTeaching the iPhone to See
34mDemystifying ML in iOS
40mDeep Learning on iOS