AI on the Edge with Computer Vision

Summary

This course provides a complete introduction on how to use the Intel® Neural Compute Stick 2 (Intel® NCS2) for low-power deep learning inference on edge devices. Topics covered include:

  • How to install the Intel® Distribution of OpenVINO™ toolkit and configure the Intel NCS2
  • The basics of deep learning vision applications and model topologies
  • How to create computer vision applications in Python* using Intel NCS2 devices

By the end of this course, students will have practical knowledge of how to use the Intel NCS2 to:

  • Analyze model performance with the included performance tools
  • Deploy pretrained networks and custom networks on the Intel NCS2
  • Deploy an object detection model on a Raspberry Pi* board

The course is structured around seven weeks of lectures and exercises. Each week requires up to three hours to complete. The code examples are implemented in Python, so familiarity with the language is encouraged (you can learn along the way).

Prerequisites

Python programming

Calculus

Linear algebra

Hardware Required

Intel NCS2

Raspberry Pi 3 Model B board or newer

Week 1

Get an introduction to the Intel NCS2. Topics include:

  • A comparison of the differences between traditional computer vision and deep learning
  • A review of the Intel® AI Portfolio including hardware and tools
  • An overview of edge inference with Intel® Movidius™ technology
  • An introduction to the Intel Distribution of OpenVINO toolkit
Week 2

See how to install the Intel NCS2. Topics include:  

  • Installation steps for the Intel Distribution of OpenVINO toolkit
  • An overview of existing pretrained models and samples that work with the toolkit
Week 3

Learn how to deploy an image classifier model on the Intel NCS2. Topics include:

  • Define an image classification model and explore a few popular image classification topologies
  • A deeper look into the Intel Distribution of OpenVINO toolkit and learn to create and deploy your first image classifier
Week 4

Learn how to deploy an object detection model on the Intel NCS2. Topics include:

  • Define an object detection model and explore a few popular object detection topologies
  • Convert and deploy a pretrained YOLO* v3 model on the Intel NCS2 using the Intel Distribution of OpenVINO toolkit
Week 5

See how to profile deep learning models using the Deep Learning Workbench. Topics include:

  • Understand the capabilities of the Deep Learning Workbench
  • Learn to install the Deep Learning Workbench directly on your system or using Docker* software
  • Profile your first deep learning model using the Deep Learning Workbench
Week 6

Learn how to deploy custom models on the Intel NCS2 using the Intel Distribution of OpenVINO toolkit. Topics include:

  • Understand what a custom model is and when to use one
  • Go through the end-to-end training and inference workflow for a custom model on the Intel NCS2
  • Implement your first custom layer using the toolkit
Week 7

Review how to deploy an object detection model on a Raspberry Pi board. Topics include:

  • Reasons to use a low-powered embedded board
  • Compare development and deployment modes of the Intel Distribution of OpenVINO toolkit
  • Install the toolkit on a Raspberry Pi board and run an object detection model