AI for Mobile and IoT Devices: TensorFlow Lite
summary
TensorFlow Lite converts TensorFlow models into a compact runtime that runs on Android, iOS, Linux and microcontrollers and claims to power 2 billion devices. The demo runs five on-device tasks in parallel—two body‑part segmentation models, real‑time matching, dynamic time warping and video encode—entirely on device and produces shareable slow‑motion dance videos. MobileNet V1 benchmarks at 83 ms on CPU and 15 ms on GPU, DSP delegation yields up to an 8.3x speed-up, and a smart mirror uses a high‑fidelity model with 400 facial landmarks.
AI-written summary · about
chapters (4)
- 0:00Start
- 11:55Conversion is sometimes hard
- 18:04Android Neural Network API standardizes acceleration
- 20:15Delegate support for Qualcomm Hexagon DSP
related talks
30mOn-device machine learning in mobile and web apps
34mTensorFlow Lite
39mAI/ML on mobile/android with TensorFlow
14mOptimize your TensorFlow Lite models
42mWhat's New with TensorFlow?
27mMachine Learning on the edge using TensorFlow Lite
27mMachine Learning and Mobile
29mFrom TensorFlow to ML Kit: power your Android application with ML