AutoML Vision and Edge TPU: Bringing TensorFlow Lite models to edge devices

Description

AutoML Vision from Google Cloud automates the process of vision recognition model training with Google’s state-of-the-art technologies such as meta learning (learning2learn) and neural architecture search (NAS) technology combined with transfer learning, data augmentation, and so on. You can just upload the training data with labels, press a train button, and wait for 24 hours to get a ML model with the cutting-edge accuracy. AutoML Vision now supports exporting of TensorFlow Lite model for prediction on smartphones, small devices (such as Rasberry Pi), and Edge TPU, the hardware accelerator developed by Google. So you can bring the power of AutoML to those edge devices and achieve significantly shorter latency on image classification.

Kaz Sato shows you how the whole process works. Without any ML expertise, you can easily train the image recognition model on the cloud, export the TensorFlow Lite model, and use it at small devices with short latency and low power consumption.