The NVIDIA Jetson Nano is a single-board computer (SBC) based on the Tegra X1 processor. In this tutorial, we show you how to use the jetson-inference tools to train an existing deep neural network to recognize different objects. The Nano is an affordable way to get started with Edge AI on an embedded system. It is capable of running NVIDIA’s CUDA language, which allows us to parallelize many machine learning tasks for faster execution. Frameworks that support CUDA, such as TensorFlow and PyTorch, work well on the Jetson Nano.
To start, we demonstrate the built-in demo of identifying a dog in real time using a webcam. From there, we move to training our own deep neural network.
For most use cases, you would want to train networks using a larger, more powerful computer, like a desktop or server. However, we can use transfer learning to train an existing network to perform similar tasks, including identifying images. We’ll use the Jetson capture tool to create a basic training set of images and use them to train a model. We can then use that model to identify images from a webcam.
Product Links:
NVIDIA Jetson Nano -
5V, 4A Power Supply -
32 GB microSD Card -
Raspberry Pi Camera Module V2 -
Related Videos:
Getting Started with NVIDIA Jetson Nano Part 1: Setup -
Intro to Edge AI: Machine Learning + IoT -
Learn more:
Maker.io -
Digi-Key’s Blog – TheCircuit
Connect with Digi-Key on Facebook
And follow us on Twitter
0 Comments