Computer vision has become increasingly important in industrial applications, serving product line management, stock control, and safety monitoring functions. However, utilizing computer vision at the edge of a network poses challenges, particularly regarding latency and reliance on mixed networks or cloud resources. To address this, Microsoft CEO Satya Nadella introduced the concept of “the intelligent edge,” bringing cloud-native tools and services to devices within networks.
While Microsoft has provided tools to containerize Azure Cognitive Services and deliver them through Azure IoT Edge, there remains a need for a solution for custom edge implementations. Containers have emerged as an ideal deployment method for edge software, with Kubernetes and service meshes offering an agnostic platform for code deployment. In this context, the KAN (KubeAI Application Nexus) project was created as an open-source solution hosted on GitHub.
KAN aims to simplify the development and management of machine learning applications on Kubernetes at scale. It provides an environment for running code on edge hardware, aggregating data from locally connected devices, and leveraging pre-trained machine learning models for insights. KAN also offers a monitoring and management portal and a low-code development environment for on-premises or cloud-based Kubernetes systems.
Notably, the KAN management portal serves as a control and monitoring interface but not as the data endpoint. It integrates with Azure Edge and AI services like Azure IoT Hub and Azure Cognitive Services, providing deeper integration when hosted on Azure. Getting started with KAN requires a Kubernetes cluster with Helm support, and Azure users can leverage Azure Kubernetes Service (AKS) for a simplified setup.
Once KAN is installed, users can build applications on the KAN portal by attaching compute devices, such as NVIDIA Edge hardware or Azure Stack Edge. KAN supports various devices running on Kubernetes clusters or Azure Edge devices. The platform also facilitates testing using Azure VMs as test devices, creating digital twins to ensure edge systems are running as expected. Industrial IP cameras are supported, and KAN enables many-to-many processing, allowing multiple applications to work with camera feeds.
Building machine learning applications with KAN involves selecting device architecture and acceleration technologies. KAN recommends using accelerated devices, such as GPUs or NPUs from NVIDIA and Intel, for safety-critical edge applications. KAN offers a node-based graphical design tool to build “AI skills,” connecting camera inputs to models and transforming/filtering outputs. Data can be exported to other applications and services, enabling customized workflows.
Once applications are built and tested, KAN simplifies packaging and deployment to target devices through the portal. Although currently limited to deploying to one device at a time, KAN aims to support deployments to multiple devices in the future. This simplifies the delivery of machine learning applications to Kubernetes systems or Microsoft’s Azure IoT Edge runtime container host, providing a centralized view of all deployments.
KAN draws inspiration from the canceled Azure Percept solution, aiming to simplify edge AI deployments with low-code tools. By adopting a similar approach to the Percept developer experience, KAN combines IoT tooling concepts with features from Microsoft’s Power Platform, enhancing the ease of building and deploying machine learning applications.
In conclusion, KAN streamlines developing and deploying machine learning applications for computer vision at the network edge. With its focus on Kubernetes and its support for various computing devices, KAN provides a platform for experimental and large-scale edge AI implementations. By simplifying the process, KAN opens up possibilities for solving challenges through edge machine learning efficiently and effectively.
Check out the GitHub link and Reference Article. Don’t forget to join our 22k+ ML SubReddit, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more. If you have any questions regarding the above article or if we missed anything, feel free to email us at Asif@marktechpost.com
Niharika is a Technical consulting intern at Marktechpost. She is a third year undergraduate, currently pursuing her B.Tech from Indian Institute of Technology(IIT), Kharagpur. She is a highly enthusiastic individual with a keen interest in Machine learning, Data science and AI and an avid reader of the latest developments in these fields.