Unlocking Data-Driven Apps with TensorFlow
summary
Shows how to build a street‑camera vehicle counter by running TensorFlow Lite inference on a Raspberry Pi with a USB camera, after training models in Google AutoML Vision using ~5,500 car images and smaller datasets for bikes and taxis. Reports that 10‑second photo intervals captured only ~25% of vehicles so switched to continuous video analytics, highlights cloud training time differences (CPU/GPU ~24h vs TPU <1h), and recommends using Coral/TPU for edge acceleration while noting AutoML requirements and Google Cloud trial credits.
AI-written summary · about
chapters (15)
- 0:00Welcome
- 1:47AI in production
- 5:59Competitive edge with AI
- 7:03TensorFlow on mobile
- 9:52AI tools in Google Cloud
- 10:57Developer adoption of AI
- 13:07Energy optimization with AI
- 15:13AI solutions platform
- 16:18AutoML: simplified training
- 23:48TPU hardware accelerators
- 27:05AI opportunities for businesses
- 28:12Demo: Raspberry Pi data capture
- 37:52Demo: real-time video analysis
- 40:44Try the tech
- 41:04Q&A
related talks
38mReal-world Machine Learning with TensorFlow and Cloud AI
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
47mPractical Intro to OpenCV & TensorFlow Lite
27mMachine Learning on the edge using TensorFlow Lite
51mCloud AI: How to Get Started Injecting AI Into Your Applications
30mTensorFlow for Mobile Heroes
59mML and the IoT: Living on the Edge
32mAI for Mobile and IoT Devices: TensorFlow Lite