Advanced on-device ML made easy
summary
MediaPipe introduces three tools — Tasks, Model Maker, and Studio — that package end-to-end on-device pipelines and cross-platform deployment. The preview ships five tasks (image classification, object detection, hand landmarks, gesture, text); Model Maker trains in ~30 seconds on ~500 images to produce a custom gesture model, the demo Rock/Paper/Scissors model exports and deploys to Android, web, and Python, and test runs show about 80% test accuracy for the example dataset. A Colab and web demo validate the pipeline and live webcam recognition.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:30Challenges in on-device ML
- 4:03MediaPipe Tasks Model Maker
- 6:57Hand gesture recognition workflow
- 8:29MediaPipe Tasks in Android
- 9:58Customizing models with Model Maker
- 11:49Off-the-shelf solutions in Studio
- 13:07MediaPipe Tasks in Python
- 15:10Deploying on the web
- 17:59Rock Paper Scissors model training
- 22:07Testing custom models
- 23:14Custom models in Python
- 24:47Rock Paper Scissors demo
- 25:10Recap of Model Maker
- 25:52Q&A
related talks
40mOn-Device Machine Learning for Android with MediaPipe
42mOn-Device Machine Learning with Google AI Edge
48mImage Recognition and Object Detection with MediaPipe on A…
36mKeynote! The opportunity for Android devs with AI and ML
15mMediaPipe Web: Bringing cross-platform AI tech to the browser
12mCross platform computer vision made easy with Model Maker
10mMachine Learning for mobile
12mML Kit: Turnkey APIs to use on-device ML in mobile apps