As I noted in an earlier article, edge computing, 5G and computationally-intensive artificial intelligence (AI) applications quickened the cloudification of network architectures spanning the data center, network core, and edge. This foundation will be key to infusing AI, machine learning (ML) and advanced analytics into the operational fabric of platforms, such as mobile, wireline, enterprise/private and cloud networks, the devices and data they connect, and the workloads and use cases they enable.
We recently highlighted successful examples of AI and ML applications in network operations and automation, and now I will share several examples of real-world deployments integrating AI, ML and analytics with the most promising and often problematic workload: video.
Infusing Network Silicon with AI Acceleration
The emerging edge environment will require a silicon foundation delivered for the unique requirements of Edge implementations, open frameworks and a robust ecosystem focused on developing commercial solutions. Before we dive into early use cases, I believe it’s important to highlight some of the key technologies that support visual and AI workloads at the edge.
Growth of video and AI applications has motivated the use of workload accelerators to deliver the desired cost- and power-optimized performance, particularly at the network edge. Our philosophy is to enhance the Intel® Xeon® CPU architecture with enhancements for accelerating workloads that are pervasive in the infrastructure. One example is the integration of Intel Deep Learning Boost (Intel DL Boost) with Vector Neural Network Instructions (VNNI) and 2nd generation Intel Xeon processors to support high performance AI inferencing at the edge for content delivery network (CDN), video analytics and other visual cloud workloads.
The Visual Cloud Accelerator Card – Analytics (VCAC-A) from our partner, Celestica, provides high-density offload acceleration of media transcode algorithms for up to 24 streams of video decode and inference on 1080p30 video in either AVC or HEVC format. This card frees up the edge CPU considerably and provides greater headroom for other applications and services. The latest version of the VCAC-A now includes integration with the Open Network Edge Services Software (OpenNESS) for easier integration into the network edge infrastructure. OpenNESS is an open source multi-access edge computing (MEC) software toolkit that enables highly optimized Edge platforms to on-board and manage applications and network functions across any type of network.
You’ll see how these important technologies work with open source toolkits and projects as well as reference pipelines and frameworks to support AI-driven applications across different edge deployment models.
Computer Vision, Open Visual Cloud and Analytics Reference Architectures
For most of our projects that require AI analytics for visual workloads, the Open Visual Inference and Neural Network Optimization (OpenVINO™) toolkit offers developers a software application programming interface (API) for accelerators, such as Intel DL Boost and VCAC-A technologies, in computer vision applications. This toolkit also preserves their software investment over multiple generations of CPUs and accelerators.
It's also helpful to have reference solutions for various visual workloads and use cases, which is where Open Visual Cloud comes into play with high performance, end-to-end sample pipelines for media, analytics, graphics and immersive media. These pipelines include Content Delivery Network (CDN) Transcode, Intelligent Ad Insertion, Video Conferencing, Interactive Ray Tracing and Smart City Traffic Management built on key Visual Cloud workloads: decode, inference, rendering and encode. The open source software stacks presented for the Open Visual Cloud are provided under the FFmpeg and GStreamer frameworks as ready-to-use Docker images and Dockerfiles to help implement cutting edge Visual Cloud services for advanced video distribution and processing. In addition, the pipelines are fully optimized for cloud native and network edge deployment on commercial-off-the-shelf CPU architecture.
Furthermore, Intel Select Solutions for Media Analytics provides a head start to build specific Visual Cloud reference pipelines with pre-verified hardware and software configurations from Intel’s ecosystem of partners. These reference solutions support real-time, high-throughput analytics on video data, streaming or otherwise, to enable live, AI-based ad insertion, and smart city solutions. The newest version of this Intel Select Solution now includes 8-bit integer precision (INT8) support with Intel DL Boost technology, delivering up to 2.5x performance improvements, compared to FP32 precision across the range of Media Analytics pipelines. INT8 and Intel DL Boost are supported by Intel Distribution of OpenVINO toolkit on the 2nd Generation Intel Xeon Scalable processor family.
While we are still in the early days of edge computing and AI deployments, let me share examples of industry collaborations that utilize many of the toolkits and reference solutions epitomized earlier.
China Mobile’s OneNET Edge Platform for Industrial Manufacturing
China Mobile’s OneNET platform supports “cloud-edge collaboration” by blending edge computing, cloud computing, AI, 5G and other technologies to provide enterprises with a comprehensive IoT architecture. The company introduced Multi-Access Edge Computing servers with the VCAC-A to deliver high computing power and DL inference capabilities for machine vision applications. One deployment focuses on intelligent manufacturing applications to improve accuracy of automated gas meter production inspection and achieve greater productivity and efficiency. The VCAC-A card is based on a single Intel Core™ processor and 12 Intel Movidius™ Myriad™ X Vision Processing Units (VPUs) to support high-density video processing capabilities.
Vision Computing Targets Crime Prevention for Retail Environments
There are millions of closed-circuit televisions deployed worldwide, but most recordings are only used after an event, like a crime or accident, has occurred. Intel and its partner, DeepSight AI Labs Pvt Ltd, have developed a computer vision solution that retrofits CCTVs with AI, video analytics and high-powered edge computing to support real-time monitoring, detection and response to criminal activity. The solution is designed using Intel Xeon processors, OpenNESS and OpenVINO.
Enabling Nokia Airframe for Video Transcoding at the Edge
We are engaged with partners to apply analytics and AI to visual workloads, especially around video streaming and smart city deployments. We partnered with Nokia to enable edge video transcoding that supports AI-driven, ad insertion in streaming video content with Nokia’s AirFrame openEDGE platform and the VCAC-A. Listen to a recent Intel Chip Chat Network Insights podcast interview with Mike Moore from Nokia to learn more.
Lenovo Re-Imagines Edge Services
Lenovo is re-imagining edge deployments with services with the latest 2nd generation Intel Xeon Scalable processors with Intel DL Boost and Intel Select Solutions for Media Analytics. Lenovo has demonstrated that its network Edge solutions will accommodate the feature pipelines intended to serve the markets of Live Artificial Intelligence (AI) Based Ad Insertion, Audio and Video (A/V) Synchronization, and Smart City Applications.
Edge Computing is fast becoming the epicenter of innovation, and the growth of data at the edge is fueling AI, analytics and video applications. My eyes light up every time I see a new AI application and begin dreaming about the things we can achieve with our partners and customers. If you share my excitement, I encourage you to bring your knowledge or level up your skills through open source projects, the Intel Network Builders University or the new Intel Edge AI of IoT Developers Nanodegree Program to name a few. I look forward to working with you to herald the new era of AI and analytics at the edge.
Connect at @rgadiyar or LinkedIn and visit https://networkbuilders.intel.com/
Notices & Disclaimers
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.
Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit www.intel.com/benchmarks.
Your costs and results may vary.
Intel technologies may require enabled hardware, software or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
 Performance and cost per stream results are based on testing as Jan 9, 2020 and may not reflect all publicly available security updates. Configurations and benchmark details are at https://builders.intel.com/docs/networkbuilders/media-analytics-solution-brief.pdf. For more complete information about performance and benchmark results, visit www.intel.com/benchmarks.
All data points are using the FFMPEG software framework. Inference Models are available from the Open Model Zoo project at https://github.com/opencv/open_model_zoo.git, checkout OpenVINO 2019_R3, 2020.1. Input Video are available on request.
- Object Detection: Mobilenet-SSD , video : person-bicycle-car-detection_1920_1080_2min.mp4, 24fps @2Mbps
- Face Recognition: face-detection-adas-0001 and face-reidentification-retail-0095 , video : face-demographics-walking_2min.mp4 , 50fps @ 10Mbps
- Car Classification: vehicle-detection-adas-0002 and vehicle- attributes-recognition-barrier-0039 , video : car-detection_1920_1080_2min.mp4, 30fps @12Mbps