Introduction to On-Device Machine Learning for Android with MediaPipe
summary
Runs full ML inference on-device using MediaPipe Tasks and TFLite models rather than cloud calls, with concrete code for image classification, object detection, hand/face landmarks, segmentation, audio classification (YAMNet) and text embedding. Shows model bundling (TFLite or task files), hardware delegates (CPU/GPU), running modes (live stream vs still), and model retraining via MediaPipe Model Maker. Claims 15 of 20 planned features are available, face mesh tracks 478 landmarks, diffusion image generation works locally in ~10–20s over 20 iterations.
AI-written summary · about
chapters (13)
- 0:00Introduction to on-device ML
- 1:11ML basics for developers
- 2:54Why MediaPipe on Android
- 5:22MediaPipe Solutions overview
- 8:12Image classification demo
- 15:55Object detection explained
- 18:02Hand landmarking and gestures
- 21:50Face landmarking and AR
- 23:54Image segmentation techniques
- 26:21Audio classification basics
- 33:09Natural language processing
- 37:01Generative ML with MediaPipe
- 40:53Future and resources
related talks
40mOn-Device Machine Learning for Android with MediaPipe
42mOn-Device Machine Learning with Google AI Edge
32mAdvanced on-device ML made easy
48mImage Recognition and Object Detection with MediaPipe on A…
36mMaking intelligent Android apps
47mMyths and Reality of Android Machine Learning
43mWhat’s new in Android machine learning
39mMachine Learning on Android Demystified