NVIDIA recently released its edge-to-cloud platform that could play a key role in the development of smart and safe cities worldwide.
NVIDIA evolved the GPU from a 3D graphics chip to a computer brain at the epicenter of the industry’s most promising endeavors such as autonomous vehicles, virtual reality and more. Now the company has launched its edge-to-cloud AI platform, NVIDIA Metropolis, which paves the way for smarter and safer cities through deep learning.
a&s Asia recently talked to Deepu Talla, VP and GM of the Tegra business at NVIDIA, to learn more about this new platform and the company’s plans for Asian markets.
a&s: In your view, how can deep learning benefit security and non-security applications?
Talla: Smart and safe cities need AI. There are approximately 500 million cameras deployed in the world today for applications such as law enforcement, traffic management and retail analytics. That number is expected to double by 2020. That’s one billion cameras, generating 30 billion frames per second of rich video data.
There’s no way for humans to analyze this scale of information. Humans currently monitor only a fraction of the captured video, with most stored on disks for later review. Traditional methods of video analytics just aren’t up to the task, so most of that rich video data is wasted.
AI and deep learning is solving problems across industries as diverse as healthcare, robotics, self-driving cars, financial services, and more. An AI video analytics system can extract meaningful insights in real time from enormous volumes of data and generate accurate results that operators can rely on to make cities smarter and safer.
a&s: Why is the GPU in a better position to run deep learning algorithms compared to the CPU?
Talla: GPUs are the platform of choice for deep learning for very practical reasons. The first is productivity. Deep learning workloads are computationally intensive and highly parallel, which means they’re well-suited to NVIDIA’s GPU architecture. GPUs make it possible for data scientists to iterate faster, creating solutions that are more accurate and robust in a shorter amount of time.
The second reason is the ecosystem. NVIDIA’s platforms are compatible with all of the major deep learning frameworks, so customers can leverage the rich ecosystem of solutions offered by AI city partners around the world.
The third reason is the platform. NVIDIA offers the only platform for the development, optimization and deployment of deep learning solutions - from the edge to the cloud - on a single computing architecture.
a&s: Can you elaborate on your latest GPU technologies/platforms/solutions? Are they for edge devices or servers?
Talla: At the GPU Technology Conference last month, we announced our NVIDIA Metropolis edge-to-cloud platform.
High-performance deep learning inferencing happens at the edge with the NVIDIA Jetson embedded computing platform, and through servers and data centers with NVIDIA Tesla GPU accelerators. Rich data visualization is powered by NVIDIA Quadro professional graphics. The entire edge-to-cloud platform is also supported by NVIDIA's software development kits, including JetPack, DeepStream, and TensorRT.
Considering where video is captured and the levels of connectivity and network infrastructure across urban areas worldwide, it’s clear that an edge-to-cloud platform is critical. AI is required at the edge for systems such as parking entrances, autonomous security robots and police cars. These need real-time insights and may not have the connectivity necessary to stream video to a centralized area for processing.
AI-enabled on-premises servers are important for places such as airports, roadway intersections, and retail environments, where a local area network exists and context across multiple cameras in both time and space is important.
Across wide geographic areas and thousands of cameras, data needs to be aggregated and analyzed so higher-level conclusions can be drawn. In these cases, cloud computing is necessary.
a&s: Can you provide some use cases where they are employed?
Talla: Global leaders in public safety and video surveillance, as well as start-ups focused on improving efficiency of city resources and operations, are using this powerful platform.
○
Traffic Management: City planners can automate traffic studies and measure parking structure activity to better manage costly parking assets.
○
Retail Analytics: Marketers and store planners can collect and leverage valuable data on customer behavior in stores for merchandising and store planning.
○
Intelligent Machines: Users of all sorts can make intelligent machines the core of their vision systems by putting superhuman vision capabilities into autonomous indoor and outdoor ground-based robots, drones, and other devices.
○
Law Enforcement: NVIDIA works with many companies that build video search tools for the security/surveillance market to complete tasks such as locating lost children in busy urban environments in minutes; analyzing thousands of hours of video to quickly bring criminals to justice; and capturing critical evidence to solve drug trafficking cases.
○
Campus safety: Companies and schools are employing video analytics that use machine learning to identify and understand normal activity, and generate alerts when anomalies occur, such as fires or violence.
a&s: What is your strategy for Asia?
Talla: We currently have more than 50 AI City partners, with several based in Asia. They include Dahua, Hanwha Techwin, Hikvision, SenseTime, Tiandy,
Umbo CV, Uniview, and others. With the release of our Metropolis platform, we’re continuing to expand our ecosystem and seeing more companies interested in developing smart and safe city applications.
We’re also focused on driving awareness around the power of AI and GPUs specifically for this space through a variety of programs. This includes the NVIDIA Deep Learning Institute, our worldwide platform to cultivate deep learning developers, as well as hosting workshops across Asia to promote GPU deep learning-based intelligent video analytics.