On-Device Machine Learning for Android with MediaPipe
summary
Demo shows on‑device MediaPipe pipelines that run real models not just slides: image generation via diffusion with repeatable seed and ~20 iterations, object detection with bounding boxes, pose landmarker with 33 keypoints, hand landmarker 21 points and face landmarker 468 points, interactive segmentation and gesture pipelines made from multiple models, and a lightweight LLM using Gemma 2B for local text tasks. Samples include runnable model files, TaskFile pipelines, and measured runtimes on phones.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:10Run inference on device
- 6:39MediaPipe vision tasks
- 10:29Setup image classification
- 15:45Object and landmark detection
- 20:44Customize models
- 23:52Generative AI on device
- 32:21Run LLMs locally
- 38:19Resources and next steps
related talks
42mOn-Device Machine Learning with Google AI Edge
32mAdvanced on-device ML made easy
48mImage Recognition and Object Detection with MediaPipe on A…
08mDemo: Gemma on-device with MediaPipe and TensorFlow Lite
47mMyths and Reality of Android Machine Learning
19mDeploying ML models to mobile devices
39mMachine Learning on Android Demystified
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML