Introducing Google Coral: Building On-Device AI
summary
Google Coral delivers an on-device AI platform built around the Edge TPU (4 TOPS, ~2 W) and two dev forms: a $150 Dev Board (Debian-based Mendel OS, snap-off SOM for $115 production modules) and a tiny USB Accelerator for Linux/Raspberry Pi. It runs 8-bit quantized TensorFlow Lite models, uses a compiler to target the TPU, and reports real workloads like ~48 FPS MobileNet object detection (comparable to a GTX 980 system) and live transferable learning via a teachable-machine demo.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 0:35Why Coral
- 2:17Key Factors
- 4:30What is Coral
- 8:48Coral Components
- 10:04Coral Devboard
- 13:04HTTPU
- 15:04Mobile Net Performance
- 16:10Dev Board
- 19:19GPIO
- 19:46Deployment
- 20:58Linux support
- 21:24Hardware support
- 22:21Software components
- 24:24Python API
- 28:12Image Classification
- 33:09What is Transfer Learning
- 34:24Coding Transfer Learning
- 35:20Teachable Machine Demo
- 38:44User Case
- 39:53Workflow
- 41:46Gravitylink
- 42:29Applications
- 43:32Contest
related talks
13mCoral: Expanding the edge AI landscape
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
47mProduction ready EdgeAI with Eclipse ioFog and Google Coral
16mIntroducing Coral Dev Board Micro
06mTop AI and ML announcements
55mEdgeAI - AI in the Palm of Your Hand
42mOn-Device Machine Learning with Google AI Edge
59mML and the IoT: Living on the Edge