Bringing Computer Vision models to the Intelligent Edge

I would like to share with you the new guide “Bringing Computer Vision models to the Intelligent Edge (with Azure IoT Edge) - A guide for developers and data scientists”.

We view the future of computing in the intelligent cloud and intelligent edge.

• The intelligent cloud is ubiquitous computing, enabled by the public cloud and artificial intelligence (AI) technology, for every type of intelligent application and system you can envision.
• The intelligent edge is a continually expanding set of connected systems and devices (that gather and analyze data - close to your users, the data, or both. Users get real-time insights and experiences, delivered by highly responsive and contextually aware apps. For this reason, this edge is now resorting to an ever-increasing use of AI technologies.

As the (Industrial) Internet of Things ((I)IoT) indeed continues to increasingly accelerate and businesses realize the immense benefits, the next breakthrough capability upon us - or even already here as industry trends - is to enable (I)IoT edge devices themselves to evolve towards a new world with:

• Billions of connected multi-sense devices on the intelligent edge.
• Seamless access to AI inference and insights at the edge, powered by the intelligent cloud.
• Serverless compute power leveraged at the edge.

Enabling intelligence on edge devices means enabling analytics and insights to happen closer to the source of the data, saving organizations money and simplifying their solutions. Increased functionalities and computing power available on the edge are already changing the way organizations design and build products, from intelligent construction site video surveillance, to oil rig maintenance tracking.

As such, the intelligent edge hardware ecosystem includes everything from traditional PCs, servers, and mobile devices to sensors, fixed purpose devices, and microcontrollers – if it’s got silicon and is Cloud-connected, it’s an edge hardware. Intelligent edge devices work with the “Intelligent Cloud” to seamlessly support distributing workloads like command and control (C2C), predictive maintenance, and any other AI-driven capabilities to the most appropriate node (whether it’s an edge device, a gateway device, an appliance, or the latency cloud) based on requirements around, performance, and security/compliance, etc.   

We are seeing this shift across many different industries – it is not just technology companies, but companies of all types who are digitally transforming and establishing their cloud and edge strategies by pushing boundaries with the applications and experiences they build.

Combining the intelligent cloud and the intelligent edge creates possibilities we could only have dreamed of just a few years ago. As Julia White, Corporate Vice President, Microsoft Azure, outlines:

The intelligent cloud and intelligent edge application pattern, transforms the way we can interact with digital information and further blend the physical and digital worlds for greater societal benefit and customer innovation.

In order to enable these different uses, the Microsoft Azure public cloud offers a set of dedicated, on-demand and ready-to-use services for IoT (Azure IoT Hub, Azure IoT Central, etc.), for IA (Azure Machine Learning Service, Azure Databricks (Spark ML), etc.), etc. which can be supplemented by open source Frameworks and APIs in the context of open innovation led by Microsoft, such as Azure IoT Edge for devices on the edge, Open Neural Network Exchange (ONNX) as a format exchange or runtime with respect to Deep Learning models for object recognition in frames of a video stream using an object detection model.

However, the most compelling solutions will be realized when it is easy to see how these many discreet building blocks can act as one end-to-end solution and combined to enable new (breakthrough) scenarios, with new revenue streams, incentives and new business partnerships. 

In this context, and illustrate possible ways, this guide proposes the implementation of an end-to-end solution from scratch from building and training of a computer vision model in the cloud to its use on a device on the edge.

This guide co-written with Song Duong as part of his internship in our team is intended for a developer and data scientist audience interested in jump starting with Computer Vision model on the intelligent edge.

The guide itself (under Documentation here), all the samples' code, Jupyter notebook files, as well as other files used in this guide are provided in a GitHub repo: https://aka.ms/IEdgeDevGuideSamples.

The webcast Deep Dive: Machine Learning on the Edge - Predictive Maintenance available on the Azure IoT Show provides a complementary in terms of AI at the edge (See also the related tutorial and repo on GitHub).

(2019-11-29 update: version 1.1 is now available with the addition of a 4th module dealing with MLOps and DevOps considerations and providing on that basis an end-to-end guided illustration with Azure Pipelines, as part of Azure DevOps Services. Enjoy the Tour! ;-))

Thanks, Philippe