ML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
summary
Live demos show machine learning compiled to run on Google Coral hardware with the EdgeTPU: a person‑detection + obfuscation pipeline runs at ~25 FPS (≈40 ms per frame) using EdgeTPU + TensorFlow Lite and the Python SDK on Mendel OS; a sushi classifier trained in AutoML exports to a TFLite Edge‑optimized model running in ~4 ms (200+ FPS) on the USB/Dev board; a Kyocera visual QA pipeline uses cropping, background removal and Hough rotation preprocessing to reach 95–98% accuracy, 8.9 ms latency and is deployed via Cloud export and device-side inference.
AI-written summary · about
chapters (8)
- 0:00Introduction and AI principles
- 1:47Understanding AI and Coral
- 7:24Benefits of edge computing
- 14:47Software tools and workflow
- 18:13Privacy protection demo
- 21:24AutoML and custom modeling
- 33:01Kyocera quality control case
- 43:47MLOps and Q&A session
related talks
45mIntroducing Google Coral: Building On-Device AI
47mProduction ready EdgeAI with Eclipse ioFog and Google Coral
59mML and the IoT: Living on the Edge
13mCoral: Expanding the edge AI landscape
01h 01mML and the IoT: Living on the Edge
55mEdgeAI - AI in the Palm of Your Hand
06mTop AI and ML announcements
42mOn-Device Machine Learning with Google AI Edge