Practical test: First ML experiments with Google Coral and TensorFlow Lite

By: MRT Desk

Published on:

Practical test: First ML experiments with Google Coral and TensorFlow Lite

table of contents

Some Raspberry Pi projects use machine learning to recognize objects, speech or gestures or to process data using neural networks. If you want to run “AI applications” in the form of deep or machine learning (DL, ML) in the IoT area, for example on an old laptop, a Raspberry Pi or other “single board computers” (SBC), you come across but quickly to the limits of computing power. Face or object recognition via a camera, for example, runs in slow motion on the Raspi. This may be enough for experiments in this area and a “proof of concept”, but not for practical use.

More about machine learning

But there is a remedy: Raspi, PC & Co. can be expanded with special chips for ML applications. The Google Coral USB3 stick used in the following (from € 69.90) accelerates object recognition on a Raspberry Pi with a camera from a limp 3 – 5 to over 30 images per second. In this article we take a look at the technical inner workings of the stick and use a sample project to determine what the stick can do.

But hardware alone is not enough, you also need the right software. The list for ML applications is long, usually PyTorch, Scikit, for example and especially TensorFlow. We use the latter in the lite version for this article. TensorFlow is a framework developed by Google for data stream-oriented programming. The term “tensor” could be seen as a generalization of scalar, vector and matrix. Google uses TensorFlow for speech recognition, GMail, Google Translate, Google Photos, Google Maps and of course Google Search. TensorFlow is written in Python and C ++ and, in addition to these two programming languages, also supports Go, Java / JavaScript, Haskel, Rust and a few more.

  • Access to all heise + content
  • exclusive tests, advice & background: independent, critically well-founded
  • Read c’t, iX, MIT Technology Review, Mac & i, Make, c’t photography directly in your browser
  • register once – read on all devices – can be canceled monthly
  • first month free, thereafter € 12.95 per month
  • Weekly newsletter with personal reading recommendations from the editor-in-chief

Start FREE month

Start your FREE month now

Already subscribed to heise +?

Sign up and read

Register now and read articles right away

More information about heise +

Article Source