TensorFlow for Mobile Heroes
summary
Shows a concrete mobile ML workflow: capture video frames with FFmpeg to get ~10,000 images, upload to Google Cloud Storage and train on Cloud ML Engine (MacBook Pro 21 hours, 4 CPUs 4 hours; one GPU ≈ 16 CPUs), convert and optimize models for mobile, and deploy updated models from Firebase. Demonstrates on-device examples (classification, detection, style transfer), Raspberry Pi cucumber classifier, and Magenta music models. Notes practical tradeoffs like quantize to 8-bit (4x smaller, ~4% accuracy loss) and TensorFlow adds ~12MB to APK.
AI-written summary · about
chapters (16)
- 0:00TensorFlow for mobile intro
- 1:34What is artificial intelligence and machine learning
- 4:46Real-world applications of machine learning
- 9:26Cloud vs on-device ML
- 10:50TensorFlow overview and GPU
- 13:39Training models with Google Cloud ML
- 16:27Neural networks learn patterns
- 17:10Optimizing models for mobile
- 19:58TensorFlow community resources
- 20:40TensorFlow Lite in Android Studio
- 22:25Demo: image classification
- 23:50Magenta creative AI models
- 24:54TensorFlow for iOS and Pi
- 26:59Swapping models in apps
- 27:43TensorFlow tools and learning
- 28:45Self-promotion and feedback
related talks
46mTensorFlow for Mobile Poets
43mTensorFlow for Mobile Developers
29mPushing Machine Learning into your app When TensorFlow and MLKit rule the world
27mMachine Learning and Mobile
31mTensorflow for Janitors
39mFrom TensorFlow to ML Kit: power your mobile application with machine learning
29mFrom TensorFlow to ML Kit: power your Android application with ML
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML